International Journal of Disaster Risk Reduction 112 (2024) 104809

Contents lists available at ScienceDirect

International Journal of Disaster Risk Reduction

journal homepage: www.elsevier.com/locate/ijdrr

Combining machine learning (ML) and participatory rural

appraisal (PRA) for disaster risk preparedness (DRP): Evidence
from the poorest region of Luzon, Philippines
Emmanuel A. Onsay a, b, * , Jomar F. Rabajante a
a
Graduate School, University of the Philippines Los Baños, Laguna, Philippines
b
Partido Institute of Economics, Partido State University, Goa, Camarines Sur, Philippines

A R T I C L E I N F O A B S T R A C T

Keywords: In the field of social science, disaster risk preparedness (DRP) is considered immeasurable due to
Disaster risk preparedness (DRP) its multidimensional nature, making it infamously difficult to quantify. The current measure­
Machine learning (ML) ments are costly, labor-intensive, and time-consuming. Consequently, policymakers struggle to
Participatory rural appraisal (PRA) target policies effectively when implementing disaster risk reduction management initiatives. By
Poorest region of Luzon in Philippines combining Participatory Rural Appraisal (PRA) and Machine Learning (ML) to train and test
Data science for disaster
community-based system datasets, this work proposes novel approaches to DRP in the poorest
region of Luzon, Philippines. We utilized sophisticated econometrics models along with ML
categorization methods. Through the analysis of 34 locales and 4 sectors within a disaggregation
system over 429 ensemble runs using cross-validation techniques, we then combined the results.
The Support Vector Machine (SVM) classifier achieved the highest accuracy of 91.55 % randomly
and 94.53 % within the pipeline, surpassing all other models. It also confirms the current rela­
tionship between DRP and multidimensional attributes (a total of 21 factors) in terms of corre­
lation and causation. Our work showcases the potential of ML for disaster risk prediction,
potentially reducing costs, saving labor, and optimizing time, especially in the most impoverished
areas of the Philippines. Ultimately, through extensive PRA, the outcomes have provided different
localities with tools for targeting policies in disaster risk management.

1. Introduction
Disasters have become a recurring phenomenon worldwide, affecting millions of people and causing unprecedented economic
losses [1,2]. The increasing frequency and severity of natural hazards, such as typhoons, floods, landslides, and earthquakes, have
highlighted the need for effective disaster risk preparedness and management [3,4]. According to the United Nations Office for Disaster
Risk Reduction (UNDRR), the world has experienced a significant increase in disaster-related losses over the past few decades, with an
estimated 1.23 million deaths and $3.7 trillion in economic losses between 1995 and 2015 (UNDRR, 2024). The impacts of disasters
are not limited to the immediate aftermath; they can have long-lasting effects on communities, economies, and ecosystems. The World
Bank estimates that disasters can push up to 26 million people into poverty each year, undermining efforts to achieve sustainable
development [5,6]. Furthermore, the rapidly changing climate is expected to increase the frequency and severity of extreme weather
events, exacerbating the risks faced by communities worldwide. In the face of these challenges, disaster risk preparedness has become a

* Corresponding author. Graduate School, University of the Philippines Los Baños, Laguna, Philippines.
E-mail address: [email protected] (E.A. Onsay).

https://doi.org/10.1016/j.ijdrr.2024.104809
Received 28 January 2024; Received in revised form 10 August 2024; Accepted 4 September 2024
Available online 8 September 2024
2212-4209/© 2024 Elsevier Ltd. All rights are reserved, including those for text and data mining, AI training, and similar technologies.
E.A. Onsay and J.F. Rabajante International Journal of Disaster Risk Reduction 112 (2024) 104809

critical imperative for governments, communities, and individuals. By understanding the risks and taking proactive measures to
mitigate and prepare for disasters, it is possible to reduce the loss of life, property, and livelihoods. Investing in Disaster Risk Pre­
paredness is economically and socially beneficial. [7] It can provide a significant return on investment by reducing the impact of
disasters on livelihoods and well-being. This can be achieved through various means, including investing in resilient infrastructure,
rebuilding stronger and more inclusively after disasters, and providing universal access to early warning systems ([8], 2020; [9]).
Overall, DRR can help reduce the economic and social losses from disasters, ultimately contributing to a decline in disaster-related
mortality and improved well-being.
The Philippines, in particular the Bicol Region, where the Partido district is located, has endured numerous destructive typhoons,
floods, soil erosion, and other natural and man-made calamities since time immemorial because of its geographic location (Bankoff,
2003 [10]; Lapidez, J. P., 2015). The Bicol Region is the poorest region in Luzon (PSA, 2022), and has been considered as the super
typhoon capital of the Philippines, the land of the howling winds, or the welcoming committee of natural calamity due to the countless
number of typhoons that battered this region, especially in Catanduanes, including Partido district of Camarines Sur [11,12]. Our
location, the Bicol Region, has been devastated by centuries of storms, but it has remained resilient in the face of Mother Nature’s
wrath. It has endured an unending barrage of storms since the beginning of time, with a startling 217 storms recorded since 1947. Six
out of ten super typhoons that make landfall in the Philippines do so directly over its territory. Super Typhoons Sening (Joan) in 1970,
Rosing (Angela) in 1995, Loleng (Babs) in 1998, Reming (Durian) in 2006, and Rolly (Goni) in 2020 ravaged the Bicol region with
horrific ferocity, leaving a terrible trail of destruction that is a sobering testament to the overwhelming force of nature. They are
renowned for their unmatched fury and have made history as the strongest typhoons to ever hit the Philippines. They ruled for many
years until the terrifying Super typhoon Rolly (Goni) overtook them in 2020. Within the annals of the Bicol Region, Goni, together with
Reming, Sisang, Loleng, Rosing, Anding, and Yolanda, is a monument to the power of storms in the history. It was the height of
destruction, indelibly carved in the memories of Bicolanos, who have a deep experience of typhoons, and bolstered by records from the
weather agency that broke all prior records for the strongest landfalling storms on earth (JTWC, 2020; PAGASA 2020; [13] Santos, G.
D. C., 2021; [14]).
Both urban and rural areas are being destroyed by these terrifying howlers, which are also creating major flash floods from Mt.
Isarog and churning the Lagonoy Gulf into catastrophic storm surges. The occurrence of calamities had some detrimental effects,
including total or partial house destruction; damage to crops or agricultural products; shocks that affected livelihood and employment;
depletion of health and nutrition; potential breakdown of peace and order; potential increase in hunger; and disruption of psychosocial
well-being as a result of these calamities [15,16]. The municipality of Goa is one of the Partido district’s most vulnerable areas in the
face of threats and disasters (MDRRMO, 2021). In light of the aforementioned, it is essential to assess the disaster risk preparedness
(DRP) of the community, how disasters affect the prevalence of poverty, and how calamity influence DRP in this particular area in
order to develop efficient intervention strategies for dealing with calamities that will inevitably recur in the future. Because, in today’s
world, disaster risk reduction management is crucial [17,18].
Typhoons are among the most dangerous natural disasters for human life. This is due to their powerful destructive power and
increased probability of occurring, as well as additional risks brought on by their aftereffects, such as storm surge and flooding brought
on by heavy precipitation [19]. The German think tank German Watch covered 182 countries worldwide to develop the 2019 Global
Climate Risk Index Report. It produced the most affected countries by disasters covering the 20-year period from 1998 to 2017. It has
ranked the Philippines as the 5th most affected country in the list topped by Puerto Rico, Honduras, Myanmar and Haiti. The German
Watch reported that the Philippines had a total of 307 weather-related events which is the most among all countries in the top ten list
[20]. The latest Intergovernmental Panel on Climate Change Assessment Report concluded that climate change will create a new poor
between now and 2100 [21]. Poverty breeds vulnerability to disasters, and consequently, those who have the least in life risk life the
most. Based on a study by the Asian Development Bank on the economics of climate change, the country stands to lose 6 % of its Gross
Domestic Product (GDP) annually by 2100 if it disregards climate change risks [22–24].
Due to its prolonged existence, difficulty to reverse, multifaceted nature, difficulty in quantifying, laborious measurement process,
and other factors, poverty is sometimes referred to as being unmeasurable. Data gathering is an expensive and time-consuming process,
making poverty measuring a notoriously difficult assignment (World Bank, 2018). Natural disasters, particularly the typhoons,
influenced poverty in the community. (Skoufias, E., 2020; Preña, E. M. et al., 2022). The impact of natural disasters on poverty can be
measured using econometric modeling, statistical analysis, subjective and objective evaluations, and other classic methods that are still
in use today. However, the scope of these measurements is restricted to confirming established as well as potential causal links.
Additionally, they are limited to using regression analysis to analyze poverty; they are unable to evaluate multidimensional predictions
that may be useful in resolving poverty in society. Therefore, when assessing poverty and natural disasters, it is necessary to include
multiple frameworks [25–27]. Poverty impedes the growth of the economy and society at large, claim Haughton and Khandker [28].
[29] All developing countries have to deal with the persistent issue of poverty. (World Bank, 2022).
Governments in developing nations distribute limited resources to alleviate poverty, decrease risk, and mitigate disasters based on
estimates of poverty, measurements of the effects of natural disasters, and quantification of disaster risk reduction management.
Nevertheless, there are a number of problems with the current methods for reducing risk vulnerabilities and focusing on programs that
reduce poverty, including inaccurate assessments, a lack of data transparency, and other influences that affect the programs already in
place. Poverty is widespread in the Philippines, both in the rural and urban areas (PSA, 2022). We are located on the island of Luzon,
which is divided into six regions. The poorest region is Region V, or Bicol. With a poverty rate of 38.7 %, Camarines Sur is the poorest
province in the six provinces that make up the Bicol Region, where we are located (PSA, 2022). There are five districts in the Camarines
Sur, the poorest of which is the Partido district, where we currently live. Typhoons and other natural disasters, in particular, are a
common occurrence in the Partido district.

2
E.A. Onsay and J.F. Rabajante International Journal of Disaster Risk Reduction 112 (2024) 104809

The natural sciences have long attempted and succeeded in measuring the unmeasurable (Rabajante, 2016; Wu et al., 2022;
Gruijters et al., 2018; Eid, 2008). However, it can be difficult to measure, assess, and in certain cases, it is impracticable to quantify
economic difficulties and social events like poverty projections, impact of disasters, and risk reduction management, specifically for
small area configurations with varied characteristics (UN, 2012).
A vast amount of data is needed to help predict disaster risk and prepare for it. Disaggregated data that can be utilized for planning
is gathered, examined, and validated using a technology-based system known as a Community-Based Monitoring System. Communities
are permitted to be involved in the process at the local level while program implementation and impact monitoring are being carried
out (Reyes, 2014; PSA, 2022). The CBMS databases include enough data to quantify the intangible components of disaster risk
management. The study of comprehending and developing learning algorithms based on substantial datasets is known as machine
learning (ML). It is an area of artificial intelligence (AI) and data science that focuses on simulating human learning processes and
increasing accuracy over time through the use of data and algorithms [30–32]. This field will be used by the research to forecast DRP
and poverty, analyze its dimensions, clarify its complexities, and pinpoint solutions. Using the CBMS data sets, ML algorithms will
create a variety of models. The training data for the model, which will enable it to make choices or predictions for natural disasters, will
come from the CBMS datasets.
Six disaster management domains have made use of deep learning (DL), machine learning (ML), and big data. Early warning
damage, damage assessment, monitoring and detection, forecasting and predicting, post-disaster coordination and response, and long-
term risk assessment and reduction have utilized ML in this six-disaster management domain [33]. Potential uses of ML in Indonesian
forests and land fires, especially smoldering peat fires, have been investigated [34]. The [35], was passed by the Philippine Congress in
2010. Aiming to address many facets of catastrophe risk reduction and management, the law strives to develop plans and strategies as
well as implement prompt and suitable measures. The Local DRRMC is tasked with planning for disaster response and recovery as well
as any consequences for the local level. Given that they are on the front lines during emergencies, it is assumed to have the major duty
for disaster response. In order to effectively deal with the reduction and management of disaster risks, it would be quite beneficial to
look at the relationship between disaster risk reduction management and poverty incidents (Onsay et al., 2024). Despite the existence
of several studies that apply machine learning (ML) to disaster risk preparedness, none have attempted to predict disaster risk pre­
paredness in a rural area by examining typhoon occurrences, conducting a complete enumeration of all households, and utilizing
disaggregated data across various socioeconomic factors. This study aims to address this gap by employing a community-centered
approach, leveraging 21 predictors from diverse socioeconomic areas to generate reliable predictions and enhance disaster risk pre­
paredness at the community level.
A wide range of societal issues, including disaster management, can benefit from participatory rural evaluation [36]. It uses a
variety of approaches, including community risk assessments (CRAs). The adoption of climate change was facilitated by this technique.
It alludes to stakeholder-used participatory techniques for evaluating risks, vulnerabilities, and capacities in support of
community-based disaster risk reduction [37]. While utilizing PRA for disaster risk reduction has been applied by many researchers,
this work is unique because it employs PRA for 34 different barangays at disaggregated level and then combining their results. It is also
different from a typical PRA because it explores scientific and nonscientific knowledge of the community by representing all residents
at different sociodemographic clusters. There is a lack of study in the Bicol region on the use of data analytics models and data science
algorithms that integrate ML, econometric models, and participatory rural appraisal (PRA) to quantify and predict disaster risk
reduction, calamity occurrences, and poverty results. In the Philippines, very little research has been done to use ML to anticipate DRP.
In fact, no study has been conducted yet in integrating ML predictions with PRA to facilitate DRP. In our work, the outcomes of DRP
prediction using ML algorithms have been combined with participatory rural evaluation.

2. Review of literature
Disaster preparedness encompasses the expertise and readiness cultivated by governmental bodies, emergency services, commu­
nities, and individuals to effectively anticipate, respond to, and recover from the impact of potential or ongoing disasters [38–40].
Disaster risk preparedness is vital for the progress of communities. Local organizations play a crucial role in providing immediate
rescue and assistance to communities to reduce the impacts and vulnerabilities to natural calamities. It is also essential for the gov­
ernment to establish direct connections among all stakeholders actively involved in Disaster Risk Reduction, especially local in­
stitutions [41]. However, preparedness levels for disasters are generally inadequate, with children and families being particularly at
risk [3]. The livelihoods and overall well-being of impoverished and vulnerable groups are threatened not only by climate change but
also by increasingly frequent extreme events such as flooding and tropical cyclones [42]. Poverty remains one of the oldest historical
issues that persist today, impacting countries worldwide. Many economists have sought to explain the root causes of poverty and the
necessary changes. Abhijit and Dulfo engaged with the impoverished, conducting surveys and gathering data with the goal of un­
derstanding the underlying ideas and motivations behind the decisions and way of life of the impoverished. They argue that these
individuals are not irrational; in fact, they may be more rational than most of us, given their need to manage limited resources [43].
However, Tingzon et al. [44] argue that traditional methods of collecting socioeconomic data can be costly, time-consuming, and
labor-intensive. Therefore, addressing this issue that concerns policymakers could greatly benefit from the application of machine
learning techniques. Considering that the study’s focus is on the poorest region of Luzon, Philippines, it is crucial to explore the
relationship and causation between poverty and preparedness for disaster risk, along with the utilization of multidimensional pre­
dictive indicators.
Big data analytics and machine learning (ML) have proven to be beneficial in the field of health by aiding in the prediction of
conditions such as diabetes [45,46]. The vast and rich datasets available from the healthcare industry are particularly valuable for ML

3
E.A. Onsay and J.F. Rabajante International Journal of Disaster Risk Reduction 112 (2024) 104809

applications [47]. However, establishing databases for disaster risk preparedness (DRP) can be costly, time-consuming, and require the
setting of various thresholds. The utilization of ML can help in measuring and predicting results and impacts in this context (Buhat, C.
A. H. et al., 2021[48]; [49]). Considering the effectiveness of ML in the natural sciences, its application in problem-solving and op­
portunity assessment can also benefit the social sciences. Machine learning (ML) has been utilized in the response to the COVID-19
disaster [31]. It has also been applied in forecasting earthquakes, tsunamis, and pandemics [50]. Support vector machines have
been employed in assessing natural disaster risks. The study’s conclusions demonstrate how the evaluation model was constructed and
how simple and effective it is. It exhibits great generalization capacity when handling small samples [51]. To precisely predict po­
tential outcomes, support vector machines have also been employed in flood risk assessment [52]. However, the utilization of machine
learning (ML) for typhoon disaster preparedness is rarely seen, and it has not yet been implemented in the Philippines. Therefore, this
study pioneers the prediction of disaster preparedness in the poorest region of Luzon, complementing ML and PRA. The majority of
research focuses on predicting the likelihood and severity of natural disasters. This process involves predicting the path and intensity of
natural disasters such as earthquakes, hurricanes, and floods using scientific and technical approaches. Researchers aim to create early
warning systems that can alert individuals and communities in affected areas, enabling them to take appropriate precautions and
mitigate the impact of the disaster. This analysis involves considering various environmental, geological, and meteorological factors
[53–55]. Predicting disaster risk preparedness (DRP) is a crucial aspect of humanity, and it is the focus of this study.
There are several applications of machine learning (ML) in disaster risk science. Amin, M. S., & Ahn, H [56]. employed an
awareness-based teaching approach for Risk and Vulnerability Assessment. KNN, LogitBoost (LB), Boosted Regression Tree (BRT),
Nearest Shrunken Centroids (NSC), and Rotation Forest were all utilized in conjunction with the adabag (AB) base classifier to map
flood vulnerabilities [57]. Nsengiyumva and Valentino [58] used ML algorithms to address the prediction of areas susceptible to
landslides, employing Random Forest (RF), Logistic Model Tree (LMT), and Naïve Bayes Tree (NBT). In the realm of disaster man­
agement, both machine learning (ML) and deep learning (DL) have found applications in early warning and disaster detection systems.
Muhammad et al. [59] introduced a methodology for early fire detection using fine-tuned Convolutional Neural Networks (CNN) on
closed-circuit television (CCTV) security cameras. Addressing the issue of false alarms in early warning systems, Chin et al. [60]
focused on enhancing the accuracy of earthquake detection. Linardos, V. et al. [31] delved into the latest advancements in disaster
management techniques involving machine learning and deep learning, specifically for damage assessment, monitoring, and
post-disaster response. Their findings suggest that DL and ML are effective tools in disaster management, indicating a rising trend in
this field. Machine learning (ML) can be categorized into three main types: Supervised Learning/Predictive Models, which are utilized
to construct predictive models. This method forecasts missing values by leveraging other values within the dataset. Through regression
and classification, it processes a set of input data and output data to create a model that can accurately predict responses for new
datasets. Unsupervised learning is another approach used to develop descriptive models. While the outcomes of this model are un­
certain, the input dataset is known. Transactional data is commonly used in this method. By incorporating both labeled and unlabeled
data in the training dataset, semi-supervised learning and clustering techniques can be employed to achieve this (Mujumdar, A. et al.,
2019). This study applies supervised ML to predict disaster preparedness in the typhoon-prone areas along the belt and in the poorest
region of Luzon, Philippines.
Moreover, an important approach known as Participatory Rural Appraisal (PRA) comprises a set of evolving techniques and
strategies that empower local populations to plan, act, and share their knowledge about life and conditions [36]. It can be employed in
social science fields to address current challenges and explore opportunities. By engaging with villagers and supporting their capacity
to plan, organize, analyze, decide, act, solve problems, monitor, and evaluate according to their needs, we utilized PRA [61]. Recent
research has witnessed a paradigm shift, transitioning from top-down directives to bottom-up planning. Consequently, local re­
searchers now adopt a participative approach rather than imposing predetermined solutions. For instance, a participatory approach
was implemented to address flood disaster management in Thohoyandou and its surrounding areas. The primary objectives of PRA
include assessing the level of crisis awareness within local communities and examining their responses to these issues [62]. In the
southwest coastal region of Bangladesh, a Participatory Rural Appraisal (PRA) was conducted to enhance disaster resilience. Re­
searchers have delved into how various calamities have impacted rural communities, the lessons derived from these experiences, the
current challenges they face concerning disasters, and the stakeholder networks they rely on to bolster their resilience [63]. Another
study used PRA to assess the status of post-cyclone livelihood capitals and to investigate the root causes of conditions in the
south-central coastal unit of Bangladesh [64]. Eleven villages in the Municipality of Guinayangan, Quezon, Philippines, took part in
the Gender Participation in Climate Vulnerability Assessment (PVA) studies. These studies examined livelihood analysis, seasonal
charting, and other climate vulnerability aspects [65]. Thus, Participatory Rural Appraisal (PRA) plays a crucial role in assessing the
knowledge and comprehension of these communities regarding disasters, along with their responses to vulnerability and preparedness
for disasters.
The Philippine Disaster Risk Reduction and Management Act, also known as Republic Act 10121, was enacted by the Philippine
Congress in 2010. The law aims to address the four recognized components of catastrophe risk reduction and management by
establishing plans and policies, as well as implementing prompt and suitable actions. The National Disaster Risk Reduction and
Management Plan comprises the following four thematic areas: 1. Preventing and mitigating disasters; 2. Being Ready for Disasters; 3.
Emergency Management; and 4. Reconstruction and Rehabilitation after a Disaster. In the event of a disaster, anticipated results,
outputs, indicators, implementing partners, and lead agencies are identified. Disasters frequently impact the socioeconomic circum­
stances of individuals in the Philippines, emphasizing the essential role of disaster risk reduction in development. To enhance disaster
risk management in the region, predicting Disaster Risk Potential (DRP) and fostering community participation are crucial. Consid­
ering the foregoing, this study investigates a broad array of (a priori) factors influencing DRP across various sectors and locations. The
authors conducted advanced econometric modeling to establish causal links between different variables and ensure that the model

4
E.A. Onsay and J.F. Rabajante International Journal of Disaster Risk Reduction 112 (2024) 104809

inputs align with econometric principles before making ML predictions. Following the ML prediction, a PRA was established to
facilitate community involvement in problem-solving and enhancing preparedness for potential disasters.

2.1. Framework
Contextualizing risk is a useful approach for quantifying it. To determine its drivers, disaster risk preparedness (DRP) needs to be
conceived. There are 3 drivers or factors that link to risk, namely: Vulnerability, Hazards, and Adaptability [66]. The concept of
vulnerability encompasses the socioeconomic conditions of a household that are inherent and difficult to alter, encompassing social
factors like health and nutrition, as well as economic factors such as poverty status. Within health and nutrition, components include
malnutrition, mortality rates, age of the household head, and household size, while poverty status involves metrics like income
poverty, food expenses, dependency ratio, and unemployment. These variables play a crucial role in determining vulnerability status
as social and economic factors significantly influence disaster preparedness. Hazards, on the other hand, represent both natural and
human-induced disasters that have the potential to occur or have impacted the community in recent years. Natural disasters comprise
events like typhoons, landslides, mudslides, floods, droughts, and volcanic eruptions, while human disasters include occurrences like
crime and hunger. These hazards are classified as threats to household safety and societal well-being. Adaptability pertains to a
household’s capacity to adapt to changing circumstances and cope with challenges, often influenced by institutional interventions and
the presence of infrastructure. Institutional interventions encompass aspects like waste management and political participation, while
infrastructure presence includes factors such as access to safe drinking water, sanitary toilet facilities, informal dwelling conditions,
and housing in makeshift structures. These variables collectively capture various risks and are considered as a priori predictors in the
Machine Learning model. These variables have undergone empirical testing, particularly through econometric modeling like logistic
regression, which considers multiple factors and intervening variables. By incorporating these components into the ML model, a
comprehensive understanding of vulnerability, hazards, and adaptability can be attained, enabling informed decision-making and
policy development to enhance disaster preparedness and resilience within communities.
The framework of our paper was also based on Resilience of the Poor in Natural Disasters ([5]; World Bank 2016). It asserts that
disaster causes huge losses to society. Thus, disaster management and risk preparedness are necessary. We have also benchmarked with
the INFORM Risk Management (2024), IASC Datasets for Disaster Preparedness and Response (2010) and UNDAC Disaster Response
Preparedness (2017). The INFORM index is a comprehensive measure designed to pinpoint nations facing potential humanitarian
crises and disasters that could surpass their domestic response capabilities. This tool facilitates a preemptive approach to crisis and
disaster management strategies. In 2024, the Philippines is rank 29 with a high INFORM risk index of 5.3 and 1.9 index as to lack of
reliability. This rank is alarming and necessitates various interventions to mitigate risk and minimize these indices. Thus, this work is
very useful to contribute for disaster risk preparedness. The IASC Datasets for Disaster Preparedness and Response (2010) were
developed to enhance the efficacy of humanitarian responses by facilitating data exchange among national authorities and humani­
tarian organizations. To ensure the efficient management of the standardized datasets endorsed by the Humanitarian Country Team, a
universal governance model should be implemented to bolster the predictability of preparedness measures. This model involves three
pivotal roles: the guardian, sponsor, and source. In this context, the data guardian is represented by the national and local statistical
agencies, the sponsor comprises academia and researchers responsible for analysis, and the source pertains to the community and local
governmental bodies. Meanwhile, The United Nations Disaster Assessment and Coordination (UNDAC) plays a crucial role in sup­
porting countries and international organizations in responding to major disasters and emergencies. It assesses disaster-affected areas
to lessen aftermath and improve recovery. It coordinates responses and efforts to improve recovery. It manages correct information and
build capacity linkages to support authorities. It is the underlying basis of our PRAs to perform assessment, coordination,
capacity-building, and information management in local context.
We have learned that data compilation and analysis is relevant to DRP. Utilizing datasets, services and tooling, literacy, gover­
nance, and networks for disaster management is vital in mitigating hazards and enhancing preparedness of the community. Van Den
Homberg, M. et. al [66]. has devised the data Preparedness and risk Framework. Risk reduction is enhanced if it is properly managed
by the community along with the participation of various stakeholders (Van De Homberg et al., 2016). Moreover, the RA10121
Philippine Disaster Risk Reduction and Management Act (2010) served as the overall guidance in developing the framework. It em­
phasizes disaster prevention and mitigation, disaster preparedness, and disaster response in the Philippines. The Act mandates the
creation of a National Disaster Risk Reduction and Management Plan, ensuring strategic disaster response strategies at all adminis­
trative levels. Central to this structure is the National Disaster Risk Reduction and Management Council, the apex body responsible for
policy formulation and plan implementation. At the grassroots level, Local Disaster Risk Reduction and Management Offices are to be
established nationwide, fostering localized disaster preparedness and response mechanisms. Emphasizing proactive measures, the Act
underscores the significance of early warning systems, capacity building, and the seamless integration of risk reduction into devel­
opmental processes. Community engagement is paramount under this legislation, with a focus on Community-Based Disaster Risk
Reduction and Management. By empowering communities to identify risks, devise plans, and implement resilience-building measures,
the Act fosters a culture of preparedness and self-reliance. For disaster preparedness as the overarching concern of this paper, we intend
to increase the level of awareness of the community to the threats and impacts of all hazards, risks and vulnerabilities, equip the
community with the necessary skills to cope with the negative impacts of a disaster, increase the capacity of institutions, and develop
and implement comprehensive local disaster preparedness policies through ML and PRA.
In addition, poverty is a multifaceted problem [67,68], which means that a number of frameworks need to be considered when
assessing poverty [25]. The relationship between economic growth and poverty is controversial. Poverty is being affected by natural
disasters and its outcomes affect the economic conditions of the society. Therefore, the framework must take into account the various
characteristics and circumstances that people have [28]. Hazards, vulnerability, and adaptation indicators are all parts of

5
E.A. Onsay and J.F. Rabajante International Journal of Disaster Risk Reduction 112 (2024) 104809

multidimensional poverty that impact one’s preparedness for disaster risk. As a result, Haughton and Khandker’s [28] notion of the
multidimensionality of poverty will also be applied. Additionally, Rowntree’s [69] minority group theory was taken into account.
Investigations among the disadvantaged subgroups were described by this early empirical theory. According to Rowntree, while it
might not deal with the underlying causes of poverty itself, it can pinpoint the causes of basic poverty. The fact that the dataset

Fig. 1. The figure depicts the taxonomy of machine learning (ML) algorithms and participatory rural appraisal (PRA) for disaster risk preparedness (DRP) prediction
and the governing frameworks.

6
E.A. Onsay and J.F. Rabajante International Journal of Disaster Risk Reduction 112 (2024) 104809

comprises a variety of sectors and subgroups makes this theory crucial to comprehending multifaceted causes. The Amartya Sen
Capability Approach was used in light of the aforementioned. The impoverished usually have limited access to opportunities, jobs,
healthcare, and education, as well as financial resources [70]. This theory has emerged as the leading substitute for traditional eco­
nomic frameworks in understanding poverty, which is relevant to this research and has an impact on an individual’s well-being.
Regardless of their socioeconomic status, the locals are capable of participating in and handling disaster management challenges.
This innovative work integrates Participatory Rural Appraisal (PRA) and Machine Learning (ML) techniques to collect and analyze
data concurrently. By separately analyzing this data, the study generates disaster risk predictions to enhance disaster risk preparedness
within communities. This framework effectively merges community insights with data-driven forecasts, guiding research towards
practical outcomes that drive positive transformations in disaster risk management. The resulting community-centric policies aim to
strengthen resilience and mitigate vulnerabilities. Moreover, these methodologies can engage diverse stakeholders to develop tailored
approaches for disaster preparedness, fostering a comprehensive and inclusive strategy to tackle disaster risks. [7,71–79] Thus, we
propose the following taxonomy of ML algorithms that might be utilized for predicting poverty, based on the previously discussed
frameworks (Fig. 1)

3. Methods
3.1. Research design
The study used both quantitative and qualitative methods (mixed method). The quantitative part utilized a secondary data through
the community-based monitoring system (CBMS). Machine learning (ML) and econometric modeling were developed and imple­
mented. A causal-explanatory research design was employed. For qualitative part, the study utilized rural participatory appraisal
(PRA) from 253 participants of the different barangays and stakeholders. They were systematically and randomly selected. The
sampling was done to ensure elimination of bias in obtaining data as well as enable the researchers to make general inferences about
the entire barangay. Qualitative methods included participatory rapid appraisal and participatory planning to generate key inputs
from local residents and promote a sense of ownership among them. The participatory approaches were conducted in 2023 at the 34
barangays of the municipality. Both Convergent Parallel Design and Transformative Design were used to harmonize Participatory
Rural Appraisal (PRA) with Machine Learning (ML). In Convergent Parallel Design, qualitative and quantitative data are collected
simultaneously but analyzed separately, then compared and integrated to identify areas of convergence and divergence [80]. This
approach provides a comprehensive view of disaster risk preparedness (DRP), combining community insights with data-driven pre­
dictions. It reveals how local perceptions align with or contradict ML findings, enriching our understanding of disaster risks and
informing proposals to improve preparedness in the region. Transformative Design integrates qualitative and quantitative methods
within a framework focused on social justice or advocacy [81]. This approach addresses disaster risk management issues, especially in
vulnerable communities. PRA empowers local voices, while ML supplies empirical data to support advocacy for disaster preparedness
resources. Together, these methods guide research toward actionable outcomes, promoting positive change in disaster risk manage­
ment through community-centered policies.

3.2. Source of data

The Goa, Camarines Sur Community-Based Monitoring System (CBMS) provided the dataset. It is the local government unit’s data.

Table 1
Variables with descriptions utilized for machine learning (ML) predictions.

Variables Type Description

1. Household Size C The total member of the household

2. Household Head Age C Biological age of HH Head
3. Informal Dwelling N 1 (HH who are informal settlers), 0 (HH who are not living in Makeshift Housing)
4. Housing in Makeshift N 1 (HH who are living in Makeshift Housing), 0 (HH who are not living in Makeshift Housing)
5. Access to safe drinking water N 1 (HH without Access to Safe Drinking Water), 0 (HH with Access to Safe Drinking Water)
6. Access to safe sanitary toilet facility N 1 (HH without Access to Sanitary Toilet Facility), 0 (HH with Access to Sanitary Toilet Facility)
7. Malnutrition N 1 (HH with malnourished children aged 0–5), 0 (HH without malnourished children aged 0–5)
8. Mortality N 1 (HH with child death below 5 years old), 0 (HH with child death below 5 years old)
9. Hunger N 1 (HH with experienced of hunger), 0 (HH without experienced of hunger)
10. Unemployment N 1 (HH with unemployment), 0 (HH without unemployment)
11. Dependency Ratio R Ratio of the number of dependents to the total HH Labor force
12. Crime N 1 (HH with crime case), 0 (HH without crime case)
13. Political Participation N 1 (HH with members who voted last election), 0 (HH without members who voted last election)
14. Waste Management N 1 (HH without proper waste management), 0 (HH with proper waste management)
15. Typhoon N 1 (HH that experienced typhoon), 0 (HH that did not experienced typhoon)
16. Flood N 1 (HH that experienced flood), 0 (HH that did not experienced flood)
17. Drought N 1 (HH that experienced drought), 0 (HH that did not experienced drought)
18. Volcanic Eruption N 1 (HH that experienced volcanic eruption), 0 (HH that did not experience volcanic eruption)
19. Landslide/Mudslide N 1 (HH that experienced land/mudslide), 0 (HH that did not experienced land/mudslide)
20. Income Poverty N 1 (HH Living below Poverty Threshold), 0 (HH Not Living below Poverty Threshold)
21. Food Poverty N 1 (HH Living below Food Threshold), 0 (HH Not Living below Food Threshold)
22. Disaster Risk Preparedness (DRP) N 1 (HH with DRP), 0 (HH without DRP)

Note: HH = Household, N = Numeric (Nominal), R = Ratio, C = Continuous.

7
E.A. Onsay and J.F. Rabajante International Journal of Disaster Risk Reduction 112 (2024) 104809

This approach was designed to provide policymakers with the necessary data to track the local economy [82]; [null]. It is a useful tool
for determining the underlying causes of poverty and disaster preparedness as well as a guide for developing successful programs and
policies that promote community development and lower poverty [83].

3.3. Data descriptions and characteristics

There are 17 variables totaling 241,740 sizes were chosen for prediction out of 14,220 units with 240 attributes in the household
dataset, which has a total size of 3,412,800. With a total size of 12,431,055, comprising 63,749 units and 195 attributes, 7 variables
totaling 442,645 were chosen for prediction from member’s datasets. The unprocessed datasets were meticulously organized and
cleared. The authors carefully altered and programmed them to comply with the econometrics and ML presumptions and guidelines.
(Table 1). An all-in system was used to fit the model. To ascertain the causation of variables involved, bidirectional, forward selection,
and backward elimination approaches were employed. Logistic and probit regression were utilized. All econometric assumptions were
satisfied: Binary outcome of dependent variables (1 for households with disaster risk preparedness (DRP), 0 for households without
DRP); independence of observations (all predictors are independent of each other); no multicollinearity (the predictors are not highly
correlated with each other); linearity of independent variables and log odds (the log odds of the outcome are a linear combination of
the independent variables); and a large sample size. All variables were categorized, the work was reviewed, the local government
provided a benchmark, physical observation of poverty through in-person visits was conducted, and data transparency and availability
were considered. The 21 variables listed below have p-values less than 0.05, lower standard errors, and narrower confidence intervals.
They serve as a priori predictors in the ML model as they are causally linked to DRP. The overall models have Prob > chi2 of 0.0000 and
Pseudo R-squared values ranging from 0.50 to 0.70.
Disaster risk preparedness (DRP) is an important factor in a household’s ability to respond and recover from emergencies and
disasters. Researchers have developed a composite score to measure this preparedness, encompassing 13 key variables. The first
variable considers a household’s water preparedness - having access to clean drinking water in times of crisis. Food preparedness is the
second variable, evaluating if the household has stocked up on essential non-perishable items like canned goods, biscuits, and bread.
The third variable looks at matches or lighters, ensuring the household has a reliable source of fire for cooking or heating. Flashlights
and emergency lighting make up the fourth variable, providing illumination during power outages. Radio or transistor preparedness,
the fifth variable, enables the household to stay informed with battery-operated communications devices. Candle preparedness, the
sixth variable, offers an alternative light source. A well-stocked medical kit is the seventh variable, allowing the household to address
basic first aid needs. The eighth variable is whistle preparedness, which can aid in signaling for help during an emergency. Clothing
preparedness, the ninth variable, ensures the household has access to appropriate attire. Blanket preparedness, the tenth variable,
helps maintain warmth and comfort. The eleventh variable addresses battery preparedness, covering items like cellphones, flashlights,
and radios. Securing important documents, such as land titles and birth certificates, makes up the twelfth variable. Finally, the
thirteenth variable evaluates money preparedness and the availability of other life-saving items the household may need (RA11315,
2018; CBMS Network, 2018; [null]). Researchers have determined that a household scoring 7 or above on these 13 variables is
considered prepared for disaster risks. Conversely, a household scoring 6 or below is deemed unprepared. This majority threshold
allows the data to be classified into a binary outcome of “prepared” or “unprepared” for disaster risk [12].

3.4. Selection of attributes

In order to determine whether variables have a correlation and a causal relationship with the dependent variables [84], we have
utilized sophisticated econometric models. We used a variety of econometric logistic models for regression analysis in order to identify
the features of variables that affect disaster risk preparedness (DRP) [12,85,86].

Y = α + Xβ + i + μ

where: Y = logit (p) = log [p/(1- p)], p = probability of disaster risk preparedness (DRP); α = the intercept or individual effects; X =
vector of independent variables; β = vector of coefficients, intercepts, or effects; and μ = error term. We have also utilized various
probit models ([12]; Campbell, H. F., 1991; Wooldridge, J. M., 2015).

P(Y = 1 |X) = Φ (XH + β)

where: P = probability of being prepared to disasters or not, Φ = Cumulative Distribution Function (CDF) of the standard normal
distribution, and β = parameters under maximum likelihood estimation. The overall model can be expressed as: M (Y|X) = P(Y = 1|X)
= Φ (β0 + β1X). The three-feature selection algorithm—information gain evaluation, correlation attribute evaluation, and learning
subset evaluation—has been implemented since we used classification algorithms. The findings show a substantial correlation and
causation between variables and a considerable impact on DRP in various aspects. With binary outcomes, classification predictions
were applied to being prepared or unprepared. Lastly, as the output shows, five folds cross validation was used in this work. The model
in this study was estimated as follows:

DRP = β0 + β1DWI + β2HIM + β3SDW + β4STF + β5MTN + β6MTY + β7HGR + β8UET + β9DRO + β10CRI + β11POL + β12WTM +
β13TYP + β14FLD + β15DRT + β16VET + β17LME + β18INP + β19FOP + β20HOS + β21HHA + βni + μ

where: DRP = Disaster Risk Preparedness; DWI = Informal Dwelling; HIM = Housing in Makeshift; SDW = Access to safe drinking
water; STF = Access to safe sanitary toilet facility; MTN = Malnutrition; MTY = Mortality; HGR = Hunger; UET = Unemployment;
DRO = Dependency Ratio; CRI = Crime; POL = Political Participation; WTM = Waste Management; TYP = Typhoon; FLD = Flood;

8
E.A. Onsay and J.F. Rabajante International Journal of Disaster Risk Reduction 112 (2024) 104809

DRT = Drought; VET = Volcanic Eruption; LME = Landslide/Mudslide; INP = Income Poverty; FOP = Food Poverty; HOS = Household
Size; HHA = Household Head Age; β0 = the intercept; β1 = the coefficient for independent variable; i = intervening variables; and μ =
the error term.
All the variables listed have a statistically significant relationship with the dependent variable of DRP, as indicated by p-values less
than 0.05. The results of the logistic and probit regression analyses are identical. The variables can be categorized into several
causation themes: household characteristics, which include household size and age of the household head; housing conditions, such as
informal dwellings and makeshift housing; access to basic services, specifically access to safe drinking water and safe sanitary toilet
facilities; household well-being, encompassing malnutrition, mortality, hunger, and unemployment; socioeconomic factors, including
the dependency ratio, crime, political participation, and waste management; disaster experience, which covers typhoons, floods,
droughts, volcanic eruptions, and landslides/mudslides; and poverty indicators, such as income poverty and food poverty. This
classification aligns with the theoretical framework (Fig. 1) and serves as a priori predictors for ML classification (Fig. 3). These
variables were selected as predictors or classifiers for ML because their correlation and causation have been empirically proven to
affect DRP. To properly predict DRP outcomes, the set of predictors must have a causal relationship with the dependent variable. The
analysis of trends and causation can be found in the discussion under Fig. 3.

3.5. Prediction methods

We selected 11 classification algorithms for predicting disaster risk preparedness (DRP) due to their unique strengths and weak­
nesses. This variety enhances our ability to capture complex relationships within the data, leading to a more robust analysis and
helping identify the most effective model for our context. Utilizing multiple algorithms facilitates comprehensive comparisons of
predictive performance, allowing us to determine which methods work best for our dataset and research questions. This approach not
only validates findings across different models but also enhances the reliability of our conclusions. Additionally, the 21 predictor
variables may exhibit diverse characteristics, including linear and non-linear relationships. A mix of algorithms enables us to effec­
tively handle these variations, ensuring a thorough analysis. Some algorithms also provide insights into feature importance, high­
lighting significant predictors for DRP, which is invaluable for stakeholders’ designing interventions. [84,87,88] By employing a
broader range of algorithms, we increase the chances of finding a model that generalizes well to new, unseen data, critical for practical
applications in this field ([89,90]; Onsay et al., 2024). Here are the 11 classification algorithms we utilized to predict DRP outcomes.
1. Adaptive Boosting (Adaboost) combines several weak classifiers to produce a powerful classifier. It emphasizes misclassified ex­
amples while iteratively training the weak classifiers. All weak classifiers’ predictions are combined and weighted according to how
well they performed to arrive at the final prediction. AdaBoost has applications in many different fields and is efficient at tackling
drp ∑
challenging categorization tasks. ̂y = Tt=1 αt γt (x), where x represents the input sample, y represents the vector of true labels for
the training samples, and T represents the total number of weak classifiers used in AdaBoost. γt (x) represents the vector of pre­
dictions of the t-th weak classifier for the N training samples. The weight αt of the t-th weak classifier is calculated as: αt =
( )
1 1− ∃t
2 ln ∃t [91].
2. Decision Tree is a ML approach that divides input data into various classes or categories by building a tree-like model. It makes use of
a hierarchical structure made up of leaf and internal nodes, where each leaf node denotes a class label and each inside node reflects
a judgment made in response to a certain feature. In order to optimize the separation of distinct classes, the algorithm learns by
recursively splitting the input data according to the feature values. To produce the most illuminating decision rules, it chooses the
optimal feature and splitting criterion at each internal node. In the decision tree, every leaf node corresponds to a distinct class
label. Let ∁ represent the class label connected to a certain leaf node. The class label linked to the leaf node that is achieved by
drp
following the decision tree’s route depending on the input features, given an input vector x, is the prediction: ̂ y = ∁ [92]
3. Gaussian Naïve Bayes is straightforward yet efficient approach for probabilistic classification. It is a probabilistic approach called
Gaussian Naïve Bayes classification for binary outcomes is based on the assumption of feature independence and Gaussian dis­
tribution for continuous features. It entails calculating the variance and mean of every feature for every class in the training set. In
order to categorize a new instance, the method applies the Bayes theorem to determine the posterior probability of each class given
the observed feature values. The anticipated class is subsequently determined by allocating the highest posterior probability class.
When dealing with binary classification jobs where the features are continuous and may be represented by a Gaussian distribution,
drp drp
this method works well and is simple to apply. ̂
y (Ci |x ) = y (ci ) y (x| ci )
y (x) where ̂
y (Ci |x ) uses the Bayes theorem to determine the
posterior probability of each class given the observed features [90,93]
4. Gradient Boost is an effective approach for binary classification. It builds a collection of weak models that optimize an objective
function by repeatedly fixing each other’s errors. It has been demonstrated that this method works quite well for making precise
(
drp drp
predictions about binary outcomes [94]. ̂ y y = (1|x ) = 1+exp1 − L(v) where ̂
y (y = (1|x ) applies a [95–97] sigmoid function to

convert the ensemble prediction into class probabilities. 1 + exp − L(v) represent the ensemble forecast by adding the weighted
prediction, v the total number of models in the ensemble, and the initial to final prediction at log-odds of the target variable.
5. Extremely Randomized Trees (Extra Trees) belongs to a decision tree family ensemble learning approach. It uses random feature
subsets and random thresholds to split nodes into numerous decision trees. By adding more randomness to the tree-building
process, Extra Trees reduce the likelihood of overfitting in comparison to regular decision trees. Extra Trees provide strong and
accurate predictions for both regression and classification tasks by combining the predictions from several trees.

9
E.A. Onsay and J.F. Rabajante International Journal of Disaster Risk Reduction 112 (2024) 104809

(
drp ∑T drp
y
̂ y = (1|x ) = T1 t=1 δt (x) where ̂
y (y = (1|x ) Calculates the average probability, T is the total number of trees, δt (x) is the

prediction on a current tree based on feature x of input data [98].

6. Linear Discriminant Analysis (LDA) a technique for classification that divides data into distinct classes using linear combinations of
characteristics. A lower-dimensional space is projected onto the data, and class separability is maximized. For classification tasks,
LDA is useful because it attempts to minimize dimensionality while maintaining the discriminatory information across classes,
particularly in cases when the classes are well-separated (Torkkola, K [99].
7. Logistic is an effective approach for binary classification problems. A logistic or sigmoid function is used to model the connection
between the input features and the binary target variable. When using logistic regression, the computer generates predictions based
on a predetermined threshold by estimating the odds that the target variable will belong to each class. Logistic regression is able to
classify new instances based on feature values by learning the coefficients for each feature by fitting the model to the training data
through an optimization procedure [90].
8. K-Nearest Neighbor is a straightforward approach for classification that uses the class labels of a new instance’s closest neighbors in
the training data to determine the class label for that instance. It determines how far the new instance is from each training instance,
chooses the K closest neighbors, and then uses a majority vote to assign the class name. Since KNN is non-parametric, it doesn’t
make any assumptions regarding the distribution of the underlying data. It is simple to comprehend and put into practice, although
it can vary depending on the distance metric and K that are chosen (Guo, G. et al., 2003).
9. Kernel SVM is a strong binary classification algorithm that can also be used for multi-class classification. The input data is trans­
formed into a higher-dimensional feature space using a method known as the kernel trick, after which it looks for the best hy­
perplane to maximally divide the classes [100].

10. Random Forest performs well and finds explanatory variables. As one of the most appropriate ML methods for variable
importance assessment, the Random Forest algorithm finds nonlinear relationships between explanatory and dependent variables
drp ∑ drp
[101]. ̂y = 12 tv=1 γv (c) where ̂ y is the predicted value of the target variable; t is the total number of decision trees in the Random
Forest; γ v (c) represents the prediction of the v-th decision tree for the input features. It is one well-liked ensemble learning approach
for classification tasks. It aggregates the separate results of several decision trees to provide predictions. Random Forest is renowned
for its capacity to manage noisy input, high-dimensional data, and feature interactions. It lowers the chance of overfitting while
producing strong and precise predictions. It also has the ability to estimate feature importance, which provides insights into the
features that have the greatest influence on the classification process.
drp
11. Support Vector Machine(SVR) is an ideal separation hyperplane between two classes is built by it. ̂ y = ( g, j) + q subject to yk
( ) ( ) drp
= g,jk − q ≤ ϑ; g, jk + q − yk ≤ ϑ where ̂ y is the predicted value; j represents the input variables; g is the weight vector; q is the
( )
bias term; yk is the actual value of the target variable for the k-th sample; ϑ is the maximum allowable deviation or error; yk = g, jk −
q ≤ ϑ is the first constraint ensures that the predicted value does not deviate more than ϑ from the actual value on the positive side; and
( )
g, jk + q − yk ≤ ϑ is the second constraint ensures that the predicted value does not deviate more than ϑ from the actual value on the
negative side. It is a strong algorithm that determines the best hyperplane to divide the data into distinct classes. In the training stage,
support vector classification (SVC) finds the examples that are closest to the decision boundary or hyperplane. By maximizing the
margin—the space between the hyperplane and the support vectors—it finds the ideal hyperplane. To address some misclassifications,
SVC takes into account both correctly categorized cases and those that fall within a specific margin known as the soft margin. SVC uses
the decision boundary’s edge to categorize new instances when generating forecasts. The classes are divided by the decision border,
and an instance’s confidence level is based on how far away it is from the decision boundary. By mapping the data into a higher-
dimensional feature space using kernel functions, SVC is useful for managing non-linearly separable data. It is renowned for its
robustness against noisy data and capacity to handle high-dimensional data, and it can handle binary and multi-class classification
problems (Cortes, C. et al., 1995; Gunn, S. R., 1998)

3.6. Disaster risk preparedness (DRP) prediction modeling

Our modeling process consists of six stages: (1) gathering data; (2) preprocessing data; (3) econometric modeling and analysis; (4)
clustering; (5) ML modeling and analysis; and (6) performance assessment. We have used many training and test sets to duplicate the
experiences while implementing cross-validation [87]. Additionally, we have used statistical significance tests and variance tests of the
performance measures:
a. Data collection – We gathered the datasets using Goa, Camarines Sur’s community-based monitoring system. (3.1–3.2)
b. Data Pre-processing – We have combed through, organized, coded, and altered the datasets to meet econometric and ML as­
sumptions. We employ a range of transformations, feature scaling via normalization, and all-in, bidirectional, backward elimi­
nation, and forward selection for the first fitting of the model. (Table 1).
c. Econometric Modeling and Analysis – A total of 78 logistic and probit models (34 locales, 4 sectors, and 1 municipal) were used to
determine correlation and causation. For ML prediction, the factors that had a strong correlation with disaster risk readiness were
prioritized. (3.3).
d. Clustering – Additionally, clustering has been used to identify a few indicators required by the classification algorithm. The datasets
were divided into prepared and non-prepared categories using K-means clustering. Due to the significant correlation between
several variables, a method was used to reliably validate the input properties. (Table 1).

10
E.A. Onsay and J.F. Rabajante International Journal of Disaster Risk Reduction 112 (2024) 104809

4 ML Modeling and Analysis – Eleven classification regressors, or algorithms, have been put into practice to forecast the results of
DRP. (3.5). Two dimensions were used to carry out these procedures [46]:
i. DRP prediction using a variety of ML techniques, we randomly selected training and test sets from the enormous datasets. Next, we
described the prediction methods (generic R and Python techniques).:
ii. DRP prediction using pipeline (established criteria), we can apply the algorithms that will produce the best accuracy outcomes
because we have built pipelines or different pre-established criteria. After fitting the pipelines to the training dataset, we compare
the accuracy outcomes. The test dataset made it easier to identify and forecast which model was the most accurate (Fig. 2).
5 Performance evaluation – we have analyzed the performance of each classification algorithms. Various evaluation metrics for
classification were applied: we utilized accuracy of classification, it is the ratio of the correctly predicted outputs to the total
number of input variables, = σδ , where α is the classification accuracy, σ is total number of correct predictions, and σ = is the total
number of predictions made; the confusion matrix, it is the matrix of output and shows the complete performance of the model
(Table 2).

The matrix accuracy is α = τ+n ∂ where α is the matrix accuracy, τ is the TP (Positive, positive prediction), ∂ is the FN (Positive,
τ where ρ is the precision, τ is the TP (Positive, positive
Negative predictions), and n is the total number of households; Precision ρ = τ+φ
prediction), φ is the FP (Negative Positive predictions). It shows the correct positive results divided by the number of positive results
divided by the classifier; Recall ∁ = τ+τ ∂ where ∁ is the recall, τ is the TP (Positive, positive prediction), and ∂ is the FN (Positive, Negative
⎡ ⎤
⎢ ⎥
⎢ ⎥
⎢( ) 1 ( )⎥
predictions). It is the ratio of total correct positive results to all the relevant households; finally, the F1 = 2 ⎢ ⎥. It reveals
⎢ ⎥
⎣ 1τ + 1τ ⎦
τ+φ τ+∂

the test accuracy, simply the harmonic means between ρ and ∁. It tells how robust and precise the classifier is.

3.7. Participatory rural appraisal (PRA)

PRA makes it easier for the community and scholars to communicate and share knowledge. It aids in a deeper comprehension of the
community’s social value system, cultural customs, and indigenous knowledge. It encourages group study of the community’s ex­
periences and current circumstances, including needs, goals, and limitations. the community’s active involvement in locating,
observing, and assessing potential fixes for the issues at hand. By making use of the resources at hand, it also aids in the creation of

Fig. 2. Disaster risk preparedness (DRP) prediction model showing the training and test data and the development of ML model with pipelines.

11
E.A. Onsay and J.F. Rabajante International Journal of Disaster Risk Reduction 112 (2024) 104809

Fig. 3. Classification accuracies of ML algorithms for disaster risk preparedness (DRP) prediction for locals and sectors at random (α).

Table 2
Disaster risk preparedness (DRP) prediction confusion matrix.

Actual Values

Predicted Values DR Prepared (1) DR Unprepared (0)

DR Prepared (1) TP Values FP Values
DR Unprepared (0) FN Values TN Values

Actual Values

Predicted Values Positive (1) Negative (0)

Positive (1) TP FP
Negative (0) FN TN

action plans that solve shortage. Collaboration between community and development organizations is strengthened as a result (Calub,
B. M., 2004). We have used the village walk, base and social mapping, village transect, and timelining to obtain a picture of the overall
geographical and social conditions of the community. We have used the seasonal calendar, resource flow diagram, Venn diagram,
trend analysis, problem cause diagram, and rating of problems to obtain more detailed information on people’s livelihood systems and
a greater knowledge of community actions. To determine the villagers’ development priorities, we employed a feasible solution rating.

12
E.A. Onsay and J.F. Rabajante International Journal of Disaster Risk Reduction 112 (2024) 104809

We used the action plan matrix to collaboratively assess and create action plans (Narayanasamy, N. 2009; Calub, B. M., 2004).

4. Disaster risk preparedness (DRP) prediction through machine learning (ML) algorithms
Communities in Camarines Sur, Philippines, particularly the fourth district, known locally as Partido Area —face a variety of issues
with regard to community development, calamity occurrences, risk reduction, and other relevant issues. However, there are also other
ways in which these issues can be resolved. Machine learning (ML) and the Partido communities’ residents’ active engagement may be
used to help come up with practical, long-lasting solutions for fair and climate-proof community development. Due to its landscape,
which includes rivers, mountains, and riverine, the area is susceptible to natural calamities. Certain mountainous settlements face not
only a lack of resources and amenities but also a heightened vulnerability to natural disasters. Goa is considered a high-risk munic­
ipality due to its proximity to Mt. Isarog, the tallest forested peak in Southern Luzon, and its exposure to the Lagonoy Gulf and the
Philippine Sea, the epicenters of tropical storms. Additionally, it has villages that are vulnerable to harsh weather conditions like
drought, landslides, and flash floods.
We have put the processes into practice and used cross-validation and prudence when applying all ML algorithms. The catastrophe
risk preparedness outcomes that are backed by econometric models serve as the target value for categorization analysis. AdaBoost,
Decision Tree, Gaussian Naïve Bayes, Gradient Boost, Extra Trees, LDA, Log istic, K-Nearest Neighbor, Kernel SVM, Random Forest,
and Support Vector Machine are the eleven classifiers used in this work. The datasets were condensed into four sectors (Isarog,
Ranggas, Salog, and Poblacion) then clustered into 34 locales (Barangays). Optimizing and guaranteeing the correctness of the ML
model classifiers is achieved by disaggregating the data at different configurations. Different ensembles are helpful in this kind of study
since different localities have different properties. The entire municipality of Goa, Camarines Sur, has 11 runs of DRP predictions, 44
runs for 4 sectors, and 374 runs for 34 locals.
The performance comparison of ML classifiers for locals and sectors is shown in Table 3. The initial model that we developed
produced the outcomes. This is predicated on randomly selected training and test sets from the vast data sets (3.6). The support vector
machine classifier, based on the analysis at random state results, has the highest accuracy for the entire municipality, at 0.91554693. It
is followed by the following algorithms: Kernel SVM, at 0.90765793; Random Forest, at 0.90715793; Logistic, at 0.90022793; Gradient
Boost, at 0.89115793; K-Nearest Neighbor, at 0.88619793; and Decision Tree, at 0.8818079. The performances of LDA and Extra Trees
are nearly identical, at 0.87858793 and 0.87916793, respectively.
The algorithms with the lowest accuracy are AdaBoost (0.84645793) and Gaussian NB (0.85665793). (Fig. 3). The outcomes at
pipeline algorithms were comparable. Nonetheless, there is a rise in classifier accuracy rates. With an accuracy rating of 94.53 %, the
SVM classification has the highest rate. Random Forest comes in second with 92.82 %, followed by Kernel SVM at 91.97 % and gradient
boost classification at 91.12 % (Table 3). The performance evaluation of all ML classifiers’ accuracies for locals and sectors at random
states and pipelines is compared in Figs. 4–5.
It is suggested that among all the classifiers, SVM produces the best classification results. These findings lend credence to the limited
body of research on DRP prediction. This aligns with the views of Gunn, S. R. (1998), Ghosh, S. et al. (2019), and Bhavsar, H. et al.
(2012), who assert that SVR is a valuable tool for classification and prediction in natural and social sciences. The findings of our work
are similar to the results obtained by other studies that have used SVR in conjunction with other algorithms. To mitigate the impacts of
flood hazards, it is crucial to first identify areas prone to such vulnerabilities. Therefore, the utilization of Support Vector Machine in
conjunction with Convolutional Neural Network has been found to enhance the efficacy of flood risk assessment by leveraging superior
image analysis capabilities (Opella, J. M. A. et al., March 2019). SVM also served as the primary model for landslide hazard assessment
in Nanping City, China, with the Bayesian Optimization (BO) algorithm employed for parameter tuning. Comparative accuracy tests
indicated that the BO-SVM model outperformed the SVM model. Notably, the hazard maps generated by the BO-SVM model
demonstrated superior overall outcomes. Thus, the combination of Support Vector Machine with an additional algorithm has shown
promise in delivering enhanced results [102]. However, most research on utilizing SVM is focused on disaster management. [103] In
2020, Sun W. et al. asserted that artificial intelligence (AI) techniques are useful for processing disaster-related data to support
informed disaster management. There is limited research on utilizing the Support Vector Machine algorithm for disaster risk pre­
paredness, particularly in the Philippines. Therefore, this work is beneficial for governments, private institutions, and individuals
working in disaster management, including those in developing countries similar to the Philippines, as it can serve as a benchmark for
making further predictions. Support Vector Machines (SVMs) are regarded as effective classifiers for binary outcomes: SVMs are
effective in high-dimensional spaces because they can manage situations in which there are more dimensions than samples, guar­
anteeing precise classification. To improve generalization and decrease overfitting, Support Vector Machines (SVMs) identify a de­
cision boundary that maximizes the margin across classes. Because support vector machines (SVMs) prioritize the data points that are
closer to the decision border, they are less susceptible to the effects of outliers [104–106]. By utilizing a kernel function to implicitly
transfer data into a higher-dimensional feature space, SVMs are able to capture intricate correlations in data. It finds the global
minimum and the best decision boundary by solving a convex optimization problem. It is also faster and use less memory because they
only use a subset of support vectors. SVMs are implemented using a wide variety of libraries and algorithms to their robust theoretical
base. They have been widely studied and applied in many different contexts. Finally, SVMs provide a strong and adaptable method for
binary classification applications.
Findings show that disaster risk readiness is greatly impacted by the likelihood of a typhoon, flood, drought, volcanic eruption, and
landslide or mudslide. The likelihood of being prepared for disaster risk rises in tandem with the frequency of natural disasters. They
do, however, significantly predict disaster readiness with regard to hunger and criminality. The degree of preparedness for disaster risk
rises in conjunction with the rates of hunger and crime. Furthermore, a household’s preparedness for disaster risk is strongly predicted

13
 E.A. Onsay and J.F. Rabajante
Table 3
Performance evaluation and comparison of machine learning (ML) algorithms for classification analysis of locals and sectors at pipelines (Ω).

Barangays and Sectors AdaBoost Decision Tree Gaussian NB Gradient Boost Extra Trees LDA Logistic K-Nearest Neighbor Kernel SVM Random Forest SVM

Abucayan 0.8869 0.9222 0.8971 0.9316 0.9196 0.9190 0.9406 0.9266 0.9481 0.9476 0.9559
Balaynan 0.9118 0.9471 0.9220 0.9565 0.9445 0.9439 0.9656 0.9515 0.9730 0.9725 0.9809
Digdigon 0.8592 0.8946 0.8694 0.9039 0.8919 0.8913 0.9130 0.8990 0.9204 0.9199 0.9283
Hiwacloy 0.9078 0.9432 0.9180 0.9525 0.9405 0.9400 0.9616 0.9476 0.9690 0.9685 0.9769
Lamon 0.8842 0.9196 0.8944 0.9289 0.9169 0.9164 0.9380 0.9240 0.9454 0.9449 0.9533
Maysalay 0.8400 0.8753 0.8502 0.8847 0.8727 0.8721 0.8938 0.8797 0.9012 0.9007 0.9091
Payatan 0.8435 0.8788 0.8537 0.8882 0.8762 0.8756 0.8972 0.8832 0.9047 0.9042 0.9125
Pinaglabanan 0.8362 0.8716 0.8464 0.8809 0.8689 0.8684 0.8900 0.8760 0.8974 0.8969 0.9053
San Isidro West 0.9125 0.9479 0.9227 0.9572 0.9453 0.9447 0.9663 0.9523 0.9737 0.9732 0.9816
Scout Fuentebella 0.8976 0.9330 0.9078 0.9423 0.9303 0.9297 0.9514 0.9373 0.9588 0.9583 0.9667
Tabgon 0.9129 0.9483 0.9231 0.9576 0.9457 0.9451 0.9667 0.9527 0.9741 0.9736 0.9820
Tamban 0.8411 0.8764 0.8513 0.8858 0.8738 0.8732 0.8948 0.8808 0.9023 0.9018 0.9102
ISAROG 0.8778 0.9132 0.8880 0.9225 0.9105 0.9099 0.9316 0.9176 0.9390 0.9385 0.9469
Bagumbayan Grande 0.6848 0.7201 0.6950 0.7295 0.7175 0.7169 0.7386 0.7245 0.7460 0.7455 0.7539
Bagumbayan Pequeño 0.7801 0.8154 0.7903 0.8248 0.8128 0.8122 0.8339 0.8198 0.8413 0.8408 0.8492
Belen 0.7533 0.7887 0.7635 0.7980 0.7861 0.7855 0.8071 0.7931 0.8145 0.8140 0.8224
La Purisima 0.6971 0.7324 0.7073 0.7418 0.7298 0.7292 0.7508 0.7368 0.7583 0.7578 0.7662
Panday 0.8484 0.8838 0.8586 0.8931 0.8811 0.8805 0.9022 0.8882 0.9096 0.9091 0.9175
14

San Benito 0.8297 0.8651 0.8399 0.8744 0.8624 0.8619 0.8835 0.8695 0.8909 0.8904 0.8988
San Isidro 0.7736 0.8090 0.7838 0.8183 0.8063 0.8057 0.8274 0.8133 0.8348 0.8343 0.8427
San Jose 0.8397 0.8750 0.8499 0.8844 0.8724 0.8718 0.8934 0.8794 0.9009 0.9004 0.9087
San Juan Evangelista 0.6832 0.7185 0.6934 0.7279 0.7159 0.7153 0.7369 0.7229 0.7444 0.7439 0.7523
San Juan Bautista 0.8777 0.9130 0.8879 0.9224 0.9104 0.9098 0.9314 0.9174 0.9389 0.9384 0.9467
POBLACION 0.7859 0.8213 0.7961 0.8306 0.8187 0.8181 0.8397 0.8257 0.8471 0.8466 0.8550

International Journal of Disaster Risk Reduction 112 (2024) 104809

Buyo 0.8609 0.8963 0.8711 0.9056 0.8936 0.8931 0.9147 0.9007 0.9221 0.9216 0.9300
Catagbacan 0.8323 0.8677 0.8425 0.8770 0.8651 0.8645 0.8861 0.8721 0.8935 0.8930 0.9014
Matacla 0.8408 0.8762 0.8510 0.8855 0.8735 0.8730 0.8946 0.8806 0.9020 0.9015 0.9099
San Pedro 0.8408 0.8761 0.8510 0.8855 0.8735 0.8729 0.8946 0.8805 0.9020 0.9015 0.9099
Tagongtong 0.8829 0.9182 0.8931 0.9276 0.9156 0.9150 0.9366 0.9226 0.9441 0.9436 0.9519
RANGGAS 0.8515 0.8869 0.8617 0.8962 0.8843 0.8837 0.9053 0.8913 0.9127 0.9122 0.9206
Cagaycay 0.8813 0.9166 0.8915 0.9260 0.9140 0.9134 0.9351 0.9210 0.9425 0.9420 0.9504
Gimaga 0.9103 0.9456 0.9205 0.9550 0.9430 0.9424 0.9640 0.9500 0.9715 0.9710 0.9793
Halawigogon 0.8607 0.8961 0.8709 0.9054 0.8935 0.8929 0.9145 0.9005 0.9219 0.9214 0.9298
Maymatan 0.8567 0.8921 0.8669 0.9014 0.8894 0.8889 0.9105 0.8965 0.9179 0.9174 0.9258
Napawon 0.8353 0.8706 0.8455 0.8800 0.8680 0.8674 0.8890 0.8750 0.8965 0.8960 0.9043
Salog 0.8623 0.8976 0.8725 0.9070 0.8950 0.8944 0.9160 0.9020 0.9235 0.9230 0.9314
Taytay 0.8872 0.9225 0.8974 0.9319 0.9199 0.9193 0.9410 0.9269 0.9484 0.9479 0.9563
SALOG 0.8705 0.9059 0.8807 0.9152 0.9032 0.9027 0.9243 0.9103 0.9317 0.9312 0.9396
GOA 0.8465 0.8818 0.8567 0.8912 0.8792 0.8786 0.9002 0.8862 0.9077 0.9072 0.9155
E.A. Onsay and J.F. Rabajante International Journal of Disaster Risk Reduction 112 (2024) 104809

Fig. 4. Overall accuracies of machine learning (ML) classifiers for locals and sectors at random state (α) and pipelines (Ω).

by household size, age of the household head, mortality, and malnutrition. There is a negative correlation between disaster pre­
paredness and household size. The likelihood of being prepared for disaster risk rises with the age of the family leader. Moreover, death
and malnutrition lower preparation for disaster risk. The degree of poverty, food costs, dependency ratio, and unemployment all have a
major impact on how prepared a household is for disaster risk. There is a negative relationship between preparedness for disasters and
the income levels and food expenses associated with poverty. It is clear that preparedness for disaster risk is adversely affected by
poverty. Furthermore, disaster preparedness rises in tandem with the amount of dependency ratio, but in reverse proportion to un­
employment. Taking Political Involvement Into account and waste Management, they are important indicators of preparedness for
potential disasters. The degree of preparedness for catastrophe risk rises in tandem with improvements in political participation and
improved waste management. The last group of factors that significantly influence DRP are the type of settlement, the kind of housing,
the availability of clean drinking water, and the availability of sanitary restrooms. A household’s level of preparedness for disaster risk
falls when they are informal settlers. A household is more likely to be ready for disaster risks when they reside in temporary housing.
Furthermore, the likelihood of a household being prepared for a disaster rises when they have access to clean, safe drinking water and
sanitary restrooms. These results are consistent with the claims made by various organizations and scholars that have investigated DRP
and factors that contribute to it (Najafi, M. et al., 2015; Akter, S. et al., 2023; Nukpezah, J. A. et al., 2018; Bankoff, G. 1999 [107];
Israel, D. C. et al., 2014). Numerous markers of DRP have been recognized. Due to the accuracy of DRP prediction models, communities
and sectors with varying socio-economic characteristics and indicators based on the classification models may receive well-targeted
policies and programs for disaster management. In light of the aforementioned, participatory rural evaluation was used to comple­
ment these findings in order to improve disaster risk reduction strategies.

5. Participatory rural appraisal (PRA)

5.1. Annals of disaster
Through Participatory Rural Appraisal (PRA), the community was asked to provide their insights about the disasters they are
familiar with and have witnessed. The authors translated this narrative from the local language provided by the households. Despite
decades of storms decimating our region, the Bicol Region, has persevered in the face of Mother Nature’s fury Fig. 6. Since the dawn of
time, it has experienced an endless stream of storms; since 1947, an astounding 217 storms have been recorded. The findings concur
with Bankoff (2003) [10], Lapidez, J. P. [108], and Onsay [12]. Super typhoons that hit the Philippines directly affect its territory,
accounting for six out of ten landfalls. The Bicol region was devastated by Super Typhoons Sening (Joan) in 1970, Rosing (Angela) in
1995, Loleng (Babs) in 1998, Reming (Durian) in 2006, and Rolly (Goni) in 2020. Their awful devastation is a sobering witness to the
power of nature. Being the strongest typhoons to have hit the Philippines, they are legendary for their unrivaled ferocity. They held

15
E.A. Onsay and J.F. Rabajante International Journal of Disaster Risk Reduction 112 (2024) 104809

Fig. 5. Comparison of machine learning (ML) classifiers for locals and sectors at random state (α) and pipelines (Ω).

power for a long time until 2020, when the powerful Super typhoon Rolly (Goni) surpassed them. Goni, Reming, Sisang, Loleng,
Rosing, Anding, and Yolanda are among the historical monuments to the strength of storms in the Bicol Region. The devastation was
unparalleled, deeply ingrained in the recollections of Bicolanos, who had extensive experience with typhoons, and reinforced by
official meteorological agency data that shattered all previous records for the greatest landfalling storms on record. This result is in
consonance with JTWC (2020), PAGASA (2020), Santos, G. D. C., (2021) [13], and Masters, J [14]. In the Bicol Region, typhoons cause
landslides, floods, mudslides, and famine, among other disasters. They also cause double disasters from other hazards in the society
such as pandemic (Preña, E. M. et al., 2022; [11]). The rain seemed like arrows penetrating the very skin of animals, and they had the
terrible potential to lift coconut shells from the land. The Partido district in Camarines Sur was completely struck by the fury of nature,
with a dreadful eye scouring the area. The quiet that came after it was only a false reprieve, quickly broken by the storm’s catastrophic
winds. Fittingly, Bicolanos called this occurrence “Ballos,” a word that conjures up feelings of both wonder and terror. The approach of
the storm’s center, or the typhoon’s eye, is much anticipated by the people of Bicol. This indicates that the storm’s initial onslaught is
slowing down, allowing for a brief period of calm before the storm picks up its furious speed and wreaks havoc with howling winds,
relentless rain, and massive waves. In order to prevent internal pressure from building up in case the wicked tempest raises their roofs
and carries them away, Bicolanos carefully open the windows on the side of their homes that is shielded from the wind. Usually, during
“ballos,” the second stage of the typhoon, the longest and strongest wind gusts hit the ground. The hazard season chart was prepared by

Fig. 6. Hazards seasonal calendar of Goa, Camarines sur.

16
E.A. Onsay and J.F. Rabajante International Journal of Disaster Risk Reduction 112 (2024) 104809

the community.
Indigenous wisdom is valuable. The ability of coastal Bicolanos to predict storms much in advance of PAGASA or any other official
meteorological service makes them exceptional. The residents of Catanduanes and the eastern Partido district also employ folk weather
analysis. Based on observations of stars, constellations, cloud formations, and Calachuci fruit behavior, this forecasting system was
developed. This traditional method of predicting also considers the behavior of hens and ducks. Other factors considered are the
direction of the wind, the size of the waves, the scent of the sea, and the presence of bubbles in the water. Additional confirmation of
the forecasts comes from birds observed during other times of the day, such as dawn, noon, and night [11,109]. There, data analytics is
used. Natural disasters happen frequently and have an impact on the homes in the communities. It is necessary to use a thorough
examination of the intervening variables between natural calamities and families in order to control risk and disaster. The majority of
locals emphasize that typhoons pose the biggest threat to the area. In addition, the area is vulnerable to drought, landslides, intense
rainfall, and flash floods. The locals claim that earthquakes can also occur in this location. Typhoons are the most common natural
disaster in the region in terms of frequency of occurrence.

5.2. Risk mapping

Risk mapping is considered essential for the correct and efficient formulation and execution of the catastrophe risk reduction
management plan. This involves doing an analysis of the capacity, vulnerability, and hazard. Depending on what a given community
needs, the approach may change. The citizens of Goa, in collaboration with the PRA facilitators, started by identifying potential
hazards, determining who could be hurt and how, and evaluating potential preventative measures. It is essential to assess the risks and
choose the appropriate control methods. The locals should take part in these events since they are more knowledgeable about and have
firsthand experience with such calamities. According to the majority of the locals, a typhoon is just a mass of intense winds and rain
that moves from one location to another. Other locals are aware that a typhoon is a severe weather event marked by powerful
northwest winds that have the potential to destroy community resources and life. Some locals think that the destruction and disruption
to their way of life and daily routine is the result of a weather storm developing in the sea or ocean. Some responders emphasized that,
despite being anticipated, the flood is unavoidable. Few respondents believed that the community’s lesson and repentance for their
transgressions are acts of God. The Philippine Atmospheric, Geophysical and Astronomical Services Administration (PAGASA) has
designated Public Storm Warning Signal No. 5, designating a typhoon capable of reaching maximum sustained winds exceeding 185
km/h. The signal is anticipated to cause extensive and severe damage in the impacted areas. When it comes to the locals’ knowledge of
the signal number 5 typhoon, the most of them merely believe that it is a powerful storm that could cause significant harm to the town.
Some have simply called it a super typhoon, capable of destroying homes, causing flooding, and maybe causing storm surges. Few
locals are ignorant of the scientific classification of this typhoon warning signal, nevertheless. Many of them declare that the
devastation of farms and plantations, as well as the damage of infrastructure, are the main effects of the typhoon, followed by loss of
livelihood. Some locals also suffer from psychological and physical ailments as a result of typhoons. This weather disturbance also
affects the availability of water and energy, internet access, and communication mediums. The methods employed in this study are
consistent with the approach described by Liu, W. [110], who integrated participatory and collaborative risk mapping to bolster
disaster resilience.

5.3. Community knowledge on disaster risk reduction and management

The Philippine Disaster Risk Reduction and Management Act, Republic Act No. 10121, is unknown to nine out of ten locals. The
Climate Change Act of 2009, Republic Act 9729, is likewise unknown to nearly the same percentage of people. Similarly, very few
locals are knowledgeable of climate change and disaster risk reduction and management (DRRM). This suggests that the majority of the
population is ignorant of relevant legislation and information about disaster risk reduction and climate change. We assessed the
residents’ contingency plans in case of impending disaster. The residents’ top priority is to move their family to a safer region in the
event that PAGASA issues a public storm warning signal number five. Additionally, they get ready to evacuate, either to a neighbor’s
concrete house or to an evacuation center. In addition, they safeguard their assets and possessions to avert potential harm in the event
of calamity. Concrete home owners just stay inside the home with their family while also providing shelter to neighborhood evacuees.
Additionally, some people are taking preventative steps by using more traditional methods including brazing homes with sturdy ropes,
piling bulky objects on the roof, and similar practices. It is essential and must to be mandated to have a location or facility that can
house individuals who are transitioning as a result of an emergency or disaster. They are typically found in community structures with
enough restrooms, a kitchen for cooking, and space for people to sleep, such as municipal halls, churches, or school gymnasiums. In
terms of the evacuation area, the majority of respondents agree that it is necessary to have one, and most locals routinely evacuate to
this location whenever a disaster approaches or threatens. Additionally, the locals are aware of the location and timing of a certain
evacuation center. The majority of responders are unaware of early warning systems as a means of preparing for the onslaught of
threats. However, the Municipal LGU, the church, and the barangay council, or BLGU, are the main organizations putting this system
into practice. It is important to have readily available and easily accessible sources of knowledge on dangers and disasters in order to
reduce or eliminate risks and prevent potential tragedies. Radio and television broadcasts are the main sources of information about
these risks or calamities. Additionally, the municipal and barangay governments educate the population in a timely manner. Infor­
mation is shared with other residents via social media, word-of-mouth from friends and acquaintances, and chain messaging. However,
the majority of responders do not have access to an emergency pack that contains valuables and documents. Out of ten residents, only
approximately four prepare an emergency kit. During emergencies, contact numbers or emergency hotlines are also crucial for
facilitating communication. Nonetheless, the majority of locals assert that it is not possible to obtain the list of crucial phone numbers.
Important documents can be stored in a storage room and kept secure in case of emergency. Approximately 70 % of the inhabitants

17
E.A. Onsay and J.F. Rabajante International Journal of Disaster Risk Reduction 112 (2024) 104809

make use of this storage space. Getting an insurance coverage is another way to be ready for risks and calamities. Disaster insurance
guards against man-made disasters as well as natural hazards and disasters like typhoons, earthquakes, floods, and tsunamis that affect
homes, companies, and livelihoods. Of all those who responded, ninety percent (90 %) do not have disaster insurance. It suggests that
the majority of the people are either unaware of this policy to the fullest extent possible, are not interested in obtaining one, or lack the
funds to do so. These findings are consistent with the outcomes of Soriano, G. [111], who conducted an evaluation of Disaster Risk
Reduction Knowledge among local residents in a chosen community, demonstrating that the community members possess a strong
understanding of disaster preparedness, readiness, adaptation, and awareness, along with a moderate grasp of disaster-related in­
formation and perceptions of disaster risk. However, our Participatory Rural Appraisal (PRA) results challenge their assertion
regarding the adequacy of disaster education initiatives in the Philippines. Our PRA indicates that the disaster education efforts in the
Bicol region are insufficient. Nonetheless, indigenous knowledge plays a crucial role in disaster resilience and mitigation. The rela­
tively low incidence of fatalities and significant lower calamity impacts among Bicolanos is not solely due to formal disaster education
but also stems from their possession of indigenous knowledge, inherent disaster resilience, and adept disaster management practices
passed down through generations.

5.4. Resilience and experiences

Resilience and persistence are well-known traits of Bicolanos and Filipinos in general. This is why most people in Goa maintain
their optimism even in the wake of natural calamities. Thirty percent were able to recover after some time, and eight percent were able
to recover right away after the natural disaster, despite the fact that over half of the respondents said they were still healing from prior
disasters. After a calamity, only 13 % of people say they are unable to recover. The locals are aware of what causes hazards and
calamities. The main causes of these calamities, according to the study participants, are deforestation, heavy rainfall, strong winds, and
poverty. These findings were also published in the PRA. Some locals claim that improper trash disposal and climate change are some of
the causes of the tragedies. Some residents think that a calamity is God’s way of forcing people to repent of their sins; these can be seen
in the table below. Because they lack adequate understanding about hazards and disasters, other residents are typically unprepared
when misfortune occurs because they do not know what to do in such situations. When it comes to family members most impacted by
disasters, elderly people and children are the most affected. Nonetheless, the outcome also indicates that individuals with disabilities
and expectant mothers are less impacted by disasters. The source of funding to support the affected community’s recovery and
rehabilitation from the damages brought about by a disaster is another crucial issue, particularly given that the average damage caused
by a disaster in Goa is approximately PhP52,000, significantly more than the average income of the barangay’s residents. While most
inhabitants utilize their own money for recovery, the DSWD and remittances from family members are also important sources of
support. Some locals obtain loans from private organizations, acquaintances, or family. A number of locals frequently take out loans
from banks and other government-run lending agencies. Others turn to loan sharks and local government assistance/subsidies for post-
disaster repairs and other forms of restoration. The incidents have an impact on the residents’ jobs or means of subsistence, frequently
resulting in losses greater than what the impacted family can manage on its own. 57 % of respondents said that farmers are the ones
most impacted by the disasters, followed by fishermen (22 %). The industry in the green and blue economies has been rendered
somewhat immobile by the disasters that have caused extensive harm to extractive fisheries and agriculture. Workers in wage-earning
and construction-related occupations are also among the most impacted. Disasters also affect jobs associated to transportation and
livestock herding. In order to ensure the family’s safety inside the home, they tie ropes to brace against wind and place sand or dirt
inside an empty bag that is inserted into the roof. Notably, 23 % of the participants are not taking any action to lessen the effects of
hazards or disasters, which is another noteworthy conclusion. These locals are either powerless in the face of dangers or certain that
they are already ready for the disaster. Campaigns to plant trees and organize evacuations are also underway, and it is evident that the
community’s faith is unwavering. Many residents prepare ahead of time, but many are unsure of where to start. They also receive
conflicting information or are deterred by the cost of taking certain precautions. Even though certain activities can be costly, there are
many low-cost or even free things they can take to lessen risk and tragedy. The activities indicated in the table below are intended to
help reduce vulnerabilities and better prepare the barangay to respond to and recover from disasters, as being prepared is an ongoing
process. The most popular behavior among the population is locking up important objects and papers. Food and water that will endure
for a few days, weeks, or even months should be stored. Restoring people’s livelihoods and other economic activity is necessary as part
of the rehabilitation and recovery process after a disaster. The majority of participants claim that it takes longer than a month for them
to be able to resume their normal lives and employment. It often takes two weeks to a month on average for livelihood to fully recover
to its ideal state. Before the way of life is fully restored, it can take anywhere from two weeks to over a month. The outcomes of these
Participatory Rural Appraisal (PRA) findings are consistent with the research by Labayo, C. C. [112], Grefalda, L. B. et al. [113], and
Preña, E. M. et al. (2022), which suggest that Bicolanos exhibit resilience in the face of disasters. These studies indicate that Bicolanos
have the capability to cope, manage, and rebuild their lives following severe calamities.

5.5. Preparing for hazards and disasters

Because they are so important, reducing risk, averting hazards, and limiting the impact of disasters ought to come first. The ma­
jority of locals follow procedures or policies that lessen the negative effects of disasters. Keeping water and food—especially canned
items, noodles, and instant coffee or drinks—in storage are some of the standard procedures. Additionally, houses are supported with
materials that are strong enough to endure strong winds and ropes. Additionally, the houses’ roofs are planked with heavy materials
like dense wood, an empty sack filled with earth, or sandbags. These items are utilized to apply pressure to a roof that could be able to
withstand the typhoon’s high sustained winds. Additionally, the occupants have torches with batteries that can be changed or
recharged. Locals frequently follow weather-related updates. Additionally, they are aware of how to get to an evacuation center in case

18
E.A. Onsay and J.F. Rabajante International Journal of Disaster Risk Reduction 112 (2024) 104809

it becomes necessary, and they consult with relevant authorities to avoid any potential harm that a particular disaster may do. Planting
trees and managing waste properly are two more noteworthy practices that the locals have noticed as part of their disaster pre­
paredness. As trees absorb greenhouse gases from the atmosphere, planting trees can help stop global warming and climate change.
Appropriate waste disposal can also stop pollution of the land and water, flash floods, and the demise of marine life. The inhabitants
ought to be equipped with pertinent DRRM knowledge. The majority of them lack experience, have not received training, or possess
insufficient professional knowledge about hazards and disasters and how to mitigate them. Twenty percent of the three hundred
participants had received DRRM training. The majority of them took part in an LGU training that was funded by the PNP, NDRRMC,
BFP, and other relevant government agencies. A portion of the residents learned about DRRM through school exercises, 4Ps lectures,
and programs run by non-profit organizations. The majority of the locals who participated in trainings learned about taking pre­
ventative steps prior to, during, and following the devastating effects of natural disasters. To lower the danger of a disaster, they are
also schooled in readiness and backup procedures. They are aware of how disasters affect society as a whole and how they specifically
harm Goa. Encounters with early warning systems, first aid training, and rescue or retrieval operations have also imparted knowledge
to the residents. Participants’ opinions about the applicability of disaster risk reduction management training are evaluated. The
majority of respondents state that their preparation for a natural disaster is significantly impacted by the training. Their decision-
making to lower the risk of a disaster and heal from medical or physiological ailments is also influenced by the instruction. The
community should place a high priority on disaster risk reduction and management training, along with any plans that may be
established to serve as a roadmap for achieving sustainable development through inclusive growth and enhancing the adaptive ca­
pacities of the local populace. In order to advance people’s welfare and security in the direction of gender-responsive and rights-based
sustainable development, the barangay is determined to carry out its obligations and responsibilities more effectively while boosting
the resilience of sectors that are particularly vulnerable and maximizing chances for disaster mitigation. Based on the aforementioned,
the participants concluded that building more evacuation facilities is necessary. Next, an early warning system is being added or
improved. Acquiring a rescue vehicle or boat is imperative, as is assembling a supply of food, water, or medical supplies. Next, an early
warning system is being added or improved. Acquiring a rescue vehicle or boat is imperative, as is assembling a supply of food, water,
or medical supplies. The locals also propose that the LGU’s assistance and the creation of possibilities for livelihood could aid the
residents in their efforts to practice DRRM. Local residents have proposed various measures such as capacity building, infrastructure
repair, tree planting, and fostering community engagement with local government officials. Given the region’s vulnerability to haz­
ards, the majority of residents demonstrate strong preparedness, resilience, and a positive outlook. Bollettino, V. et al. [114], affirms
that the Bicol region stands out as a location characterized by high levels of preparedness, adaptability, coping strategies in disasters,
and the ability to recover effectively.

5.6. Disaster risk reduction and management plan via participatory approach
The DRRM plans that the locals and stakeholders produced and discussed during the PRA are listed in the tables below. Every
interested resident, regardless of age, gender, industry, or socioeconomic status, took part in the participatory planning process. The
outputs are summarized.

5.6.1. Disaster preparedness

In order to guarantee future preparedness for typhoons, the following measures should be put into practice: evacuate individuals
who are vulnerable two days before the event, with the assistance of MDRRMC officials and BLGU Tanod, as responsible agencies, and
a budget allocation of P50,000-PhP150,000; print monthly information, education, and communication (IEC) materials involving the
BLGU, PSU, and MPDC, with a budget of P10,000–20,000; conduct ten trainings within six months to a year led by the MDRRMO and
LGU, allocating a budget of P75,000–200,000 for all barangays; and build Rainwater Collection (RWC) projects in each barangay with
the participation of the LGU, BLGU, DPWH, DAR, and PSU, allocating a budget of P4,000,000 sourced from national and local gov­
ernment agencies. [115] Disaster preparedness in Bicol region is crucial for recovery and development ([null]; Preña, E. M. et al., 2022;
Bollettino, V. et al. [114].

5.6.2. Disaster prevention and mitigation

Typhoons must be prevented and their effects must be minimized by putting in place a number of precautions, including: Firstly, to
reduce crop damage and guarantee food security, start early harvesting procedures five days before to the anticipated arrival of the
typhoon. By being proactive, we may lessen possible losses and protect agricultural resources. Second, start an extensive tree-planting
initiative with the goal of planting 1000 trees in each barangay in a year. Work together with the DENR and BLGU to manage this
project and allot P20,000–P100,000 in funding. This initiative will improve environmental resilience, lessen soil erosion, and lower the
risk of landslides by increasing green cover. Next, give spine boards, vital medical supplies for emergency response, and safe means of
transporting injured people during typhoon events to every barangay. Set aside P20,000 in the budget to help pay for the acquisition of
these necessary instruments. In addition, provide a single vehicle every barangay a priority purchase over a period of one to two years.
Work with the PGO, MGO, BLGU, and LTO to assign a P350,000 budget each barangay in order to streamline the procurement process.
This will provide quick relief and evacuation by improving the effectiveness of emergency response activities. In order to hasten the
clearance of debris and blockages after a typhoon, make sure that clearing operation equipment is acquired. The processes of recovery
and restoration can be sped up by quickly addressing these risks. The results align with the recommendations of Evasco, K. E. et al.
[116], advocating for the enhancement of promotional and capacity-building initiatives concerning disaster management. They also
propose the integration of disaster management into the developmental agendas of local government units. Consequently, it is
imperative for the Local Government Unit (LGU) of Goa to offer services aimed at mitigating the impact of disasters both before and

19
E.A. Onsay and J.F. Rabajante International Journal of Disaster Risk Reduction 112 (2024) 104809

during their occurrence.

5.6.3. Disaster response

There are a few steps that need to be done to guarantee an efficient disaster response. Prioritizing the evacuation and rescue of
vulnerable groups, such as people with disabilities (PWD), children, elderly people, pregnant women, and those living in temporary
housing or informal settlements, should be the top priority for rescue operations. To guarantee their safety, these activities must to be
carried out at least two days prior to the typhoon’s arrival. To carry out these rescue operations, the relevant authorities—the Barangay
Local Government Unit (BLGU), Barangay Police, and Bureau of Fire Protection (BFP)—should work together. The resources required
for these operations would be provided by allocating a budget of P30,000 each barangay. During a typhoon, food assistance is very
important for evacuees. Working together with the Local Government Unit (LGU), BLGU, and Department of Social Welfare and
Development (DSWD) will guarantee that food supplies are distributed to people in need in a timely and appropriate manner.
Appointing a P25,000 budget for each evacuation center will help ensure that there are enough food resources available. Effective
distribution requires coordination with the Municipal Disaster Risk Reduction and Management Council (MDRRMC), the Barangay
Disaster Risk Reduction and Management Council (BDRRMC), and other pertinent organizations. First aid and other medical care
should be easily accessible both during and after the hurricane. Working together with the Red Cross and the Department of Health
(DOH) will guarantee that medical staff and supplies are available. It is imperative to respond promptly to medical emergencies and to
provide basic healthcare services. This include giving out first aid and providing further medical attention as needed. In 2015 [117],
has written that collaboration among government and non-government institutions is vital in disaster response. It was proven to be
effective in an oil spill disaster in Guimaras Is. Philippines. Concerning the situation of Goa, Camarines Sur, it should create linkages
with private institutions, academia, and national bodies to be able to further manage disaster response.

5.6.4. Disaster rehabilitation and recovery

Effective rehabilitation and recovery initiatives are essential to the reconstruction of communities and the empowerment of
impacted individuals following a disaster. It is essential to take many crucial steps in order to accomplish this. First off, it is possible to
determine the full level of destruction by quickly assessing damages in the initial days following the disaster. Through the allocation of
a P50,000 fund each barangay and the involvement of accountable authorities like Tanod and the BLGU, this assessment guarantees
prompt and focused support. It’s also crucial to start a thorough clearing effort one week to one month following the disaster. Working
together with organizations such as the DWPH, LGU, and DND and setting aside P1,000,000 for the entire municipality will make it
possible to clear trash and repair essential infrastructure. As a result, the impacted communities will have a safe space to recuperate
and rebuild.In the rehabilitation phase, organizing socio-economic seminars and trainings is essential. By providing knowledge and
resources, these initiatives promote resilience and empower communities to “build back better.” This approach ensures that recovery
efforts are sustainable and address the underlying socio-economic challenges. Participating in orientation and planning meetings
following a disaster also enables communities to take an active role in their own recovery. This phase makes communities more
resilient by encouraging a sense of solidarity and ownership and makes ensuring that reconstruction operations meet the needs and
goals of the impacted populace. Given the significance of mental health, psychological counseling services must be offered. A budget of
P12,000 per barangay, in conjunction with organizations such as PSU, BHW, RHU, and BLGU, would take care of the afflicted people’s
emotional and mental health needs. Communities can mend and rebuild their lives stronger than before by providing care and
assistance. Regarding disaster rehabilitation and recovery, the Bicol Region in particular and the Philippines in general have imple­
mented diverse recovery programs. Community engagement has been identified as a crucial element in post-disaster recovery efforts.
The findings of the Participatory Rural Appraisal (PRA) align with the conclusions drawn by Florano, E. R. [118], Soriano, G. [111],
and Preña, E. M. et al. [119] regarding the vital role of community involvement in the post-disaster recovery process.

6. Discussion and recommendations

This study used methods to forecast the sectoral (4) and local (34) levels of disaster risk preparedness (DRP) in Goa, Camarines Sur,
Philippines. The datasets were created by utilizing the municipality’s community-based monitoring program. In order to anticipate
DRP, we facilitated classification methods. We used eleven ML classifiers for the classification analysis. The categorization of families
as prepared or unprepared for the danger of disaster is the aim value. For classification analysis, the Support Vector Machine provides
the greatest accuracy rates, greater recall, precision, and K1. There is a significant causal relationship between the variables and
qualities and the DRP outcomes. There is a substantial correlation between DRP in the local community and sectors and the associated
indicators of hazards, vulnerability, and adaptability.
Given that the majority of disaster science research conducted in the nation used the traditional descriptive methodology, the
results of this work are encouraging. In order to provide more accurate and useful DRP forecasts that can be used to target policies and
programs for rural localities and sectors in disaggregated and aggregated configurations, we have blended econometrics, ML, and
participatory rural evaluation in this work. This is a promising result that can be implemented in all other Philippine towns, as well as
in all other areas and the nation at large. The nation’s current DRP measurements are based on costly surveys and censuses and are
historical in nature. These censuses and surveys cost a lot of money, take a lot of time, and need a lot of effort.
Furthermore, the government’s present analyses are primarily descriptive in nature. Because we offer diagnostic, predictive, and
prescriptive methodologies and models, the output could therefore enhance the data analytics process in the rural sector. By using the
datasets from the community-based monitoring system, this study is in line with the goals of the NDRRMC and Local Government Unit
to estimate and anticipate multidimensional-driven DRP. It means that the public will have more opportunities to reduce expenses and
get rid of non-value-adding operations related to gathering data and formulating policy recommendations for disaster recovery

20
E.A. Onsay and J.F. Rabajante International Journal of Disaster Risk Reduction 112 (2024) 104809

management. Additionally, it ensures the efficient gathering of data that will protect the researchers and enumerators in isolated
locations, risky regions, and perilous terrain where social, political, and geographic constraints are common.
We may share our ML algorithms with other organizations and use them to support policy targeting systems that are intended to
address disaster recovery (DR) management on a local level. Our algorithms have been specifically created and applied for the
community and sectors. Our efforts are in line with the sustainable development objectives, which, particularly in our area—the
poorest region of Luzon, Philippines—aim to make cities and human settlement inclusive by 2030 through the reduction of disaster risk
and the improvement of DRP.
To improve accuracy and decrease errors, deep learning (DL) techniques might be incorporated into the ML models developed in
this study for further research projects. The result should then be used as input for a policy targeting system in the municipality’s
localities and sectors to minimize expenses, labor, and time while optimizing outputs to lower risk and effectively handle disasters. DR
preparation efforts for locals with different qualities will undoubtedly be focused while reducing time, cost, and labor by using the
PRA’s output. We so conclude that PRA and ML regression are potential tools for disaster management in the Philippines, and we
strongly encourage the public to employ our models.
The results of the PRA and survey clearly show that, despite having a low income, little understanding of DRRM rules, and a dearth
of disaster supplies and equipment, Goans are highly aware of and prepared for hazards and catastrophes. Even while some households
take precautions and mitigate some of the effects of the disaster, these are determined to be insufficient and may call for more
aggressive government action to either improve household-level readiness or the government’s reaction to the needs of the inhabitants,
or both.
The most common and noticeable hazard in the area is typhoons. Precautions must be taken prior to, during, and following the
disaster. The community’s preventive or preemptive measures should be consistently practiced, which is based on the recorded replies
and actions of the community. If disaster is imminent, homeowners should prepare a pack with clothing, batteries, a flashlight, water,
canned goods, and other essentials. The locals give priority to moving their family to a safer region in the event that PAGASA issues a
public storm warning signal that requires more careful thought. Locals keep an eye on typhoon-related news on radio, television, and
social media. A radio with a rechargeable battery is utilized in the event of a power outage.
Cellphones and other devices can also be charged using power banks. Families are also urged to keep a sufficient quantity of food on
hand, including rice, noodles, canned products, and food that keeps well even without refrigeration. Additionally, homes are inspected
and repaired for damage to ensure that powerful winds and torrential rains won’t get through. Candles and kerosene lamps, radios,
flashlights, and other lighting sources are also ready. A first aid kit needs to be locked up. The authorities’ orders are complied with. To
the closest evacuation center, some locals flee. In order to minimize flooding, people also assist in maintaining the community’s canals
and drainage system.
In order to avoid mishaps, electric poles are also inspected. Residents notify the electric cooperative so that it can promptly
reinforce any weak electric post hazards they become aware of. Tree branches near homes that might fall on them should be pruned. In
addition, younger family members could receive instructions on what to do in the event of a typhoon. Residents are prepared to flee if
needed during the typhoon’s passage. As soon as they know the family members are no longer safe, they depart the residence.
Additionally, they prepare some bottled water or boil drinking water.
Staying inside a car or other vehicle during a typhoon is normally safe for the occupants, but not during flooding. Finally, it is a
routine to keep an eye on reports and weather forecasts. Regarding actions taken following the storm, homeowners inspect their home
for damage to the roof, doors, windows, and water lines, and they promptly perform any necessary repairs. To keep from becoming
sick, they boil the water before consuming it. People protect themselves against potentially sharp or pointy items by wearing shoes,
slippers, or other types of footwear. They stay away from cut electrical wires. The locals also avoid floodwaters by staying away from
them. Lastly, while always remaining in a safe location, residents clean up and remove anything that was harmed by the typhoon.
The DRRM plan that Goa officials and citizens actively participated in creating is an excellent spot to start. The plan outlines the
steps that need to be done before to, during, and following a significant disaster in the region. When the necessity arises, responsible
authorities and the necessary funding are also identified to further facilitate the plan’s implementation. Because the plan was created
by the locals, the people of Goa feel more invested in it and are therefore more likely to implement it.
The goals of the national DRRM are in line with the Goa DRRM strategy, which are as follows: Disaster Prevention and Mitigation:
lessen community susceptibility and exposure to all hazards and increase community capacities to lower risks and deal with the effects
of all hazards; Disaster preparedness is the process of raising community awareness of hazards, risks, and vulnerabilities and their
effects; giving people the tools they need to deal with the aftermath of a disaster; building institutional capacity; and creating and
implementing comprehensive national and local plans, policies, and systems for disaster preparedness; Disaster Response: reduce the
number of avoidable fatalities and injuries, meet the basic needs of the impacted population, and promptly restore basic social services;
and Disaster Rehabilitation and Recovery: restore people’s means of subsistence and the continuity of economic activity and business;
rebuild infrastructure and other public utilities; restore shelter and other buildings/installations; and help those who were harmed
physically and psychologically. By merging PRA and ML, realistic plans to handle disaster concerns might be enabled for the socio­
economic progress of our region.

7. Conclusion
This study successfully demonstrates the application of machine learning (ML) and participatory rural appraisal (PRA) to enhance
disaster risk preparedness (DRP) in Goa, Camarines Sur, Philippines. By utilizing data from community-based monitoring programs
and employing eleven ML classifiers, particularly the Support Vector Machine, we achieved significant accuracy in predicting DRP

21
E.A. Onsay and J.F. Rabajante International Journal of Disaster Risk Reduction 112 (2024) 104809

status among households. The findings reveal a strong correlation between community characteristics and DRP outcomes, indicating
that while local awareness of disaster risks is high, actual preparedness measures remain inadequate. The existing reliance on
traditional, descriptive methodologies for disaster management is insufficient, necessitating the adoption of more dynamic, predictive
approaches through ML and community-centered methods through PRA.

Ethics statement
The Partido State University and the University of the Philippines Los Baños approved the conduct of this study under NTP2023-
RP1-S1“Measuring the Unmeasurable” research project. The distribution, use, manipulation, examination, and application of the data
outputs that underwent processing were approved by the Local Government Unit of Goa, Camarines Sur, Philippines. The data analysis
and methods are carried out willingly by researchers; they are not a part of any experiment. Our human data is secondary and indirect.
The local government approved the PRA data collection, and participants gave their informed consent. Additionally, no direct human
volunteers, animal experimentation, or data gathered from social media platforms are used in this study.

CRediT authorship contribution statement

Emmanuel A. Onsay: Writing – review & editing, Writing – original draft, Visualization, Validation, Supervision, Software, Re­
sources, Project administration, Methodology, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization.
Jomar F. Rabajante: Writing – review & editing, Supervision, Software, Project administration, Methodology, Investigation, Formal
analysis, Conceptualization, Funding acquisition.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to
influence the work reported in this paper.

Data availability

Data will be made available on request.

Acknowledgements
The authors wish to express their profound gratitude to Elsevier’s International Journal for Disaster Risk Reduction (IJDRR) for
offering a platform to share their work on disaster risk preparedness in developing nations. They extend their appreciation to the
University of the Philippines Los Baños for their assistance in conceptualization and analytical support, the Partido State University for
their funding assistance, the De La Salle University’s School of Economics for their methods and reviews, and the Local Government of
Goa, Camarines Sur, for their valuable data provision. The authors deeply acknowledge Hon. Marcel S. Pan, Engr. Moriel Prado, Ms.
Lea Nonah M. Perit, Ms. Keschei Joana Villar Cañaveras, Mr. Herman B. Jungco II, Mr. Jude Zair C. Paladan, Mr. Chris Ocampo, and
Mr. Joebert Ronabio for their contributions in data provision, enumeration, authorization, and logistics. The authors are grateful to Dr.
Jason Alinsunurin, Dr. Neil Jerome Egarquin, and Dr. Jefferson Arapoc for their assistance in econometrics and mathematical analysis.
The authors express their profound appreciation to Prof. Eva Marie Aragones and Dr. Alellie Sobreviñas for their assistance on
developmental models. A special thanks is extended to Prof. Rolan Jon G. Bulao, Sir Kevin C. Baltar, Mark Rey Pardiñas, Sakura Kokok,
Spotty Kankan, Korukoy Ogik, Kinkon Ukay, Biboy Zakat, Binangon Karankan, Jakjak Nonoy, Yanyan Ilay, Kokey Kokoy, Bonbon
Buday, and Bokbok Gigantoy for their efforts in editing, curation, and logistics. Above all, to the All-Powerful God for His wisdom,
might, and knowledge. Let God be praised!

References
[1] W.W. Botzen, O. Deschenes, M. Sanders, The economic impacts of natural disasters: a review of models and empirical studies, Rev. Environ. Econ. Pol. (2019).
[2] C.I. Donatti, K. Nicholas, G. Fedele, D. Delforge, N. Speybroeck, P. Moraga, A. Zvoleff, Global hotspots of climate-related disasters, Int. J. Disaster Risk Reduc.
108 (2024) 104488.
[3] K.R. Ronan, E. Alisic, B. Towers, V.A. Johnson, D.M. Johnston, Disaster preparedness for children and families: a critical review, Curr. Psychiatr. Rep. 17
(2015) 1–9.
[4] J. Twigg, Disaster Risk Reduction: Mitigation and Preparedness in Development and Emergency Programming, Overseas Development Institute (ODI), 2004.
[5] S. Hallegatte, A. Vogt-Schilb, M. Bangalore, J. Rozenberg, Unbreakable: Building the Resilience of the Poor in the Face of Natural Disasters, World Bank
Publications, 2016.
[6] World Bank, Reducing disaster risk from natural hazards: an evaluation of World Bank support. https://ieg.worldbankgroup.org/sites/default/files/Data/
reports/ap_disasterriskreduction.pdf, 2021.
[7] S. Hallegatte, J. Rentschler, B. Walsh, Building Back Better: Achieving Resilience through Stronger, Faster, and More Inclusive Post-disaster Reconstruction,
World Bank, 2018.
[8] S. Hallegatte, J. Rentschler, J. Rozenberg, Lifelines: the Resilient Infrastructure Opportunity, World Bank Publications, 2019.
[9] UNDRR, The human cost of disasters: an Overview of the last 20 Years, 2000–2019, Geneva: UNDRR, https://www.undrr.org/publication/human-cost-
disasters-2000-2019, 2020.
[10] G. Bankoff, Blame, responsibility and agency:‘Disaster justice’and the state in the Philippines, Environ. Plann.: Nature and Space 1 (3) (2018) 363–381.
[11] R. Sarmiento, The Catandunganons—making and seeing selves through the eye of the typhoon, AghamTao 20 (2011) 68–86.
[12] E.A. Onsay, Poverty profile and health dynamics of indigenous people, Int Rev Soc Sci Res 2 (2022) 1–27.
[13] G.D.C. Santos, 2020 tropical cyclones in the Philippines: a review, Trop. Cycl. Res. Rev. 10 (3) (2021) 191–199.
[14] J. Masters, Super Typhoon Goni slams into Philippines as strongest landfalling tropical cyclone on record, Yale Climate Connections 1 (2020).

22
E.A. Onsay and J.F. Rabajante International Journal of Disaster Risk Reduction 112 (2024) 104809

[15] D.C. Israel, R.M. Briones, Impacts of natural disasters on agriculture, food security, and natural resources and environment in the Philippines (No. 2012-36),
PIDS discussion paper series (2012).
[16] A.V. De Viana, The Philippines’ typhoon alley: the historic bagyos of the Philippines and their impact, Jurnal Kajian Wilayah 5 (2) (2014) 184–216.
[17] V. Sandoval, M. Voss, V. Flörchinger, S. Lorenz, P. Jafari, Integrated disaster risk management (IDRM): elements to advance its study and assessment, Int. J.
Disaster Risk Sci. (2023) 1–14.
[18] I. Pińskwar, A. Choryński, D. Graczyk, Risk of flash floods in urban and rural municipalities triggered by intense precipitation in wielkopolska of Poland, Int. J.
Disaster Risk Sci. (2023) 1–18.
[19] G. Liu, X. Li, J. Wang, Y. Kou, X. Wang, Research on the statistical characteristics of typhoon frequency, Ocean Eng. 209 (2020) 107489.
[20] D. Eckstein, M.L. Hutfils, M. Winges, Global Climate Risk Index 2019. Who Suffers Most from Extreme Weather Events, vol. 36, 2018.
[21] O.C. Change, Intergovernmental Panel on Climate Change, vol. 52, World Meteorological Organization, 2007.
[22] ADBI Major Report Details Potential Costs of Climate Change in the Pacific, ADB, 2013. https://www.adb.org/news/major-report-details-potential-costs-
climate-change-pacific-adb.
[23] D. Dole, P. Abeygunawardena, An Analysis and Case Study of the Role of Environmental Economics at the, Asian Development Bank, 2002.
[24] D. Castells-Quintana, M. del Pilar Lopez-Uribe, T.K. McDermott, Adaptation to climate change: a review through a development economics lens, World Dev.
104 (2018) 183–196.
[25] S. Bibi, Measuring Poverty in a Multidimensional Perspective: A Review of Literature, 2005.
[26] D. Alexander, Natural disasters: a framework for research and teaching, Disasters 15 (3) (1991) 209–226.
[27] R.L. Ciurean, D. Schröter, T. Glade, Conceptual frameworks of vulnerability assessments for natural disasters reduction, in: Approaches to Disaster
Management-Examining the Implications of Hazards, Emergencies and Disasters, 2013, pp. 1–32.
[28] J. Haughton, S.R. Khandker, Handbook on Poverty and Inequality, World Bank, Washington, DC, 2009.
[29] A. Deaton, Measuring poverty in a growing world (or measuring growth in a poor world), Rev. Econ. Stat. 87 (1) (2005) 1–19.
[30] L.B.L. Santos, L.R. Londe, T.J. de Carvalho, D. S Menasché, D.A. Vega-Oliveros, About interfaces between machine learning, complex networks, survivability
analysis, and disaster risk reduction, Towards mathematics, computers and environment: a disasters perspective (2019) 185–215.
[31] V. Linardos, M. Drakaki, P. Tzionas, Y.L. Karnavas, Machine learning in disaster management: recent developments in methods and applications, Machine
Learning and Knowledge Extraction 4 (2) (2022).
[32] R. Quiliche, B. Santiago, F.A. Baião, A. Leiras, A predictive assessment of households’ risk against disasters caused by cold waves using machine learning, Int. J.
Disaster Risk Reduc. 98 (2023) 104109.
[33] R.R. Arinta, E.A. Wr, Natural disaster application on big data and machine learning: a review, in: 2019 4th International Conference on Information
Technology, Information Systems and Electrical Engineering (ICITISEE), IEEE, 2019, November, pp. 249–254.
[34] H. Riza, E.W. Santoso, A. Kristijono, D.N. Melati, F. Prawiradisastra, Machine learning application in response to disaster risk reduction of forest and peatland
fire: impact-based learning of DRR for forest, land fire and peat smouldering, Majalah Ilmiah Pengkajian Industri 14 (3) (2020).
[35] Philippine disaster risk reduction and management act, also known as republic act 10121. https://www.officialgazette.gov.ph/2010/05/27/republic-act-no-
10121/, 2010.
[36] R. Chambers, Participatory rural appraisal (PRA): analysis of experience, World Dev. 22 (9) (1994) 1253–1268.
[37] M.K. Van Aalst, T. Cannon, I. Burton, Community level adaptation to climate change: the potential role of participatory community risk assessment, Global
Environ. Change 18 (1) (2008) 165–179.
[38] B.T. Tyubee, Disaster preparedness: approaches and frameworks, in: W. Leal Filho, A. Azul, L. Brandli, A. Lange Salvia, P. Özuyar, T. Wall (Eds.), No Poverty.
Encyclopedia of the UN Sustainable Development Goals, Springer, Cham, 2020, https://doi.org/10.1007/978-3-319-69625-6_15-1.
[39] A. Wiltshire, Developing early warning systems: a checklist, in: Proc. 3rd Int. Conf. Early Warning (EWC), 2006, March, pp. 27–29.
[40] UN/ISDR, Developing early warning systems: a checklist, United Nations International Strategy for Disaster Reduction (UNISDR). UNISDR (2006).
[41] A.A. Shah, R. Shaw, J. Ye, M. Abid, S.M. Amir, A.K. Pervez, S. Naz, Current capacities, preparedness and needs of local institutions in dealing with disaster risk
reduction in Khyber Pakhtunkhwa, Pakistan, Int. J. Disaster Risk Reduc. 34 (2019) 165–172.
[42] K.S. Ochoa, T. Comes, A Machine learning approach for rapid disaster response based on multi-modal data. The Case of Housing & Shelter Needs, 2021 arXiv
preprint arXiv:2108.00887.
[43] Esther Duflo, Abhijit Banerjee, Poor Economics, vol. 619, PublicAffairs, New York, NY, USA, 2011.
[44] I. Tingzon, A. Orden, K.T. Go, S. Sy, V. Sekara, I. Weber, D. Kim, Mapping poverty in the Philippines using machine learning, satellite imagery, and crowd-
sourced geospatial information, Int. Arch. Photogram. Rem. Sens. Spatial Inf. Sci. 42 (2019) 425–431.
[45] E. Ramsahai, N. Dookeram, D. Ramsook, J.R. Rameshwar, A.B. Yearwood, A. Bachoo, K. Khan, Crime prediction in Trinidad and Tobago using big data
analytics: predictive policing in developing countries, International Journal of Data Science and Analytics 15 (4) (2023) 421–432.
[46] A. Mujumdar, V. Vaidehi, Diabetes prediction using machine learning algorithms, Procedia Computer Science 165 (2019) 292–299.
[47] A. Rodríguez-González, A. Vakali, M.A. Mayer, T. Okumura, E. Menasalvas-Ruiz, M. Spiliopoulou, Introduction to the special issue on social data analytics in
medicine and healthcare, International Journal of Data Science and Analytics 8 (2019) 325–326.
[48] C.A.H. Buhat, M.C. Torres, Y.H. Olave, M.K.A. Gavina, E.F.O. Felix, G.B. Gamilla, J.F. Rabajante, A mathematical model of COVID-19 transmission between
frontliners and the general public, Network Modeling Analysis in Health Informatics and Bioinformatics 10 (2021) 1–12.
[49] Jomar F. Rabajante, Jerrold M. Tubay, Hiromu Ito, Takashi Uehara, Kakishima Satoshi, Satoru Morita, Yoshimura Jin, Dieter Ebert, Host-parasite Red Queen
dynamics with phase-locked rare genotypes, Sci. Adv. 2 (3) (2016) e1501548.
[50] P. Purushotham, D.D. Priya, A. Kiran, Disaster analysis using machine learning, in: 2022 International Conference on Advancements in Smart, Secure and
Intelligent Computing (ASSIC), IEEE, 2022, November, pp. 1–6.
[51] J. Xiong, J. Li, W. Cheng, N. Wang, L. Guo, A GIS-based support vector machine model for flash flood vulnerability assessment and mapping in China, ISPRS
Int. J. Geo-Inf. 8 (7) (2019) 297.
[52] J.M.A. Opella, A.A. Hernandez, Developing a flood risk assessment using support vector machine and convolutional neural network: a conceptual framework,
in: 2019 IEEE 15th International Colloquium on Signal Processing & its Applications (CSPA), IEEE, 2019, March, pp. 260–265.
[53] M.A. Manivasagam, C. Ramya, R. Bhumika, S. Roshan, B. Nirmal, S. Madhusudhan, NATURAL DISASTER PREDICTION USING MACHINE LEARNING, 2023.
[54] N. Gupta, K.K. Rana, Disaster prediction and post disaster management using machine learning and bluetooth, Webology 18 (5) (2021) (ISSN: 1735-188X).
[55] T. Sharma, A. Singhal, K. Kundu, N. Agarwal, Machine learning/deep learning for natural disasters, in: Applications of Artificial Intelligence, Big Data and
Internet of Things in Sustainable Development, CRC Press, 2022, pp. 91–121.
[56] M.S. Amin, H. Ahn, Earthquake disaster avoidance learning system using deep learning, Cognit. Syst. Res. 66 (2021) 221–235.
[57] P. Prasad, V.J. Loveson, B. Das, M. Kotha, Novel ensemble machine learning models in flood susceptibility mapping, Geocarto Int. 37 (16) (2022) 4571–4593.
[58] J.B. Nsengiyumva, R. Valentino, Predicting landslide susceptibility and risks using GIS-based machine learning simulations, case of upper Nyabarongo
catchment, Geomatics, Nat. Hazards Risk 11 (1) (2020) 1250–1277.
[59] K. Muhammad, J. Ahmad, S.W. Baik, Early fire detection using convolutional neural networks during surveillance for effective disaster management,
Neurocomputing 288 (2018) 30–42.
[60] T.L. Chin, C.Y. Huang, S.H. Shen, Y.C. Tsai, Y.H. Hu, Y.M. Wu, Learn to detect: improving the accuracy of earthquake detection, IEEE Trans. Geosci. Rem. Sens.
57 (11) (2019) 8867–8878.
[61] L. Cavestro, PRA-Participatory Rural Appraisal Concepts Methodologies and Techniques, Padova University, Padova PD. Italia, 2003.
[62] N.I. Sinthumule, N.V. Mudau, Participatory approach to flood disaster management in Thohoyandou, Jàmbá: Journal of Disaster Risk Studies 11 (3) (2019)
1–7.

23
E.A. Onsay and J.F. Rabajante International Journal of Disaster Risk Reduction 112 (2024) 104809

[63] G.A. Parvin, R. Dasgupta, M.A. Abedin, M. Sakamoto, B. Ingirige, M.G. Kibria, H. Nakagawa, Disaster experiences, associated problems and lessons in
southwestern coastal Bangladesh: exploring through participatory rural appraisal to enhance resilience, Sustainable and Resilient Infrastructure 8 (sup1)
(2023) 223–236.
[64] T.B. Taher, S.H. Rahman, Impact of cyclone sidr on rural livelihood using participatory rural appraisal (PRA) tools: a case study of a coastal unit in Bangladesh,
Journal of Sustainability Science and Management 13 (1) (2018) 91–102.
[65] M.M. Rosimo, R.R. Vidallo, J.B. Dalusag, E.B. Bernardo, E. Monville-Oro, B.O. Rosales, J.F. Gonsalves, Participatory Vulnerability Assessments in Support of
Community Based Adaption, 2021.
[66] M. Van Den Homberg, J. Visser, M. Van Der Veen, Unpacking data preparedness from a humanitarian decision making perspective: toward an assessment
framework at subnational level, in: Proc. Information Systems for Crisis Response and Management (ISCRAM) Conference, 2017.
[67] A. Sen, Welfare, freedom and social choice: a reply, Recherches Économiques de Louvain/Louvain Economic Review 56 (3–4) (1990) 451–485.
[68] A. Sen, Development as Freedom, The globalization and development reader: Perspectives on development and global change, 2014, p. 525.
[69] B.S. Rowntree, Poverty: A Study of Town Life, vol. 1901, Macmillan, London, 1901, pp. 119–120.
[70] A. Sen, Valuing Freedoms. Sen’s Capability Approach and Poverty Reduction, Oxford University Press, New York, Oxford, 1999.
[71] IASC Guidelines Common Operational, Datasets (CODs) in disaster preparedness and response. https://www.humanitarianresponse.info/en/applications/
tools/category/operational-datasets, 2010.
[72] IASC Task Force on Information Management, Common Operational Datasets in disaster preparedness and response. https://interagencystandingcommittee.
org/sites, 2010.
[73] INFORM index for risk Management, Using INFORM to Assess Crisis and Disaster Risk within a Country or Region, Guidance Note, 2016.
[74] INFORM Risk, Inter-agency standing committee reference group on risk, early warning and preparedness and the European commission. https://drmkc.jrc.ec.
europa.eu/inform-index/INFORM-Risk/Results-and-data/moduleId/1782/id/469/controller/Admin/action/Results, 2024.
[75] UNDAC disaster response preparedness missions Part I strategy and framework. http://www.pacificdisaster.net/pdnadmin/data/original/UNDAC_Mission_
PartI_strategy_framwrk.pdf, 2017.
[76] UNDAC United Nations Disaster Assessment and Coordination, Field Handbook, seventh ed., 2018. https://www.unocha.org/publications/report/world/un-
disaster-assessment-and-coordination-undac-field-handbook-7th-edition-2018.
[77] UNDRR, Integrating Disaster Risk Reduction and Climate Change Adaptation in the UN Sustainable Development Cooperation Framework, UNDRR, Geneva,
2020. https://www.undrr.org/publication/integrating-disaster-risk-reduction-and-climatechange-adaptation-un-sustainable.
[78] United Nations Office for Disaster Risk Reduction (UNDRR). https://www.undrr.org/, 2024.
[79] M.J.C. van den Homberg, R. Monné, M.R. Spruit, Bridging the information gap: mapping data sets on information needs in the preparedness and response
phase, Technologies for Development 213 (2018).
[80] P.P. Bhairawa, S. Suryanto, N. Sinta, W. Ida, R. Yan, Using convergent parallel mixed methods and datasets for science, technology, and innovation policy
dynamics research in Indonesia, ASEAN Journal on Science and Technology for Development 39 (2) (2022) 61–68.
[81] D.M. Mertens, Transformative research methods to increase social impact for vulnerable groups and cultural minorities, Int. J. Qual. Methods 20 (2021)
16094069211051563.
[82] Alellie Borel Sobreviñas, The Community-Based Monitoring System (CBMS): an Investigation of its Usefulness in Understanding the Relationship between
International Migration and Poverty in the Philippines, University of Antwerp, 2017.
[83] RA11315, Community-based monitoring act of 2018. https://www.officialgazette.gov.ph/downloads/2019/04apr/20190417-RA-11315-RRD.pdf, 2018.
[84] E.A. Onsay, J.F. Rabajante, Measuring the unmeasurable multidimensional poverty for economic development: datasets, algorithms, and models from the
poorest region of Luzon, Philippines, Data Brief 53 (2024) 110150.
[85] C. Reyes, A. Tabuga, C. Mina, R. Asis, M. Datu, Chronic and Transient Poverty (PIDS Discussion Paper Series No. 2010-30), Philippine Institute for
Development Studies, 2011.
[86] Alellie Borel Sobreviñas, Examining chronic and transient poverty using the community-based monitoring system (CBMS) data: the case of the municipality of
orion, DLSU Business & Economics Review 30 (1) (2020) 111–129.
[87] E.A. Onsay, J.F. Rabajante, When machine learning meets econometrics: can it build a better measure to predict multidimensional poverty and examine
unmeasurable economic conditions? Science Talks (2024) 100387.
[88] E. Onsay, J. Rabajante, Measuring the unmeasurable through machine learning regressions and classifications: multidimensional poverty predictions in the
poorest region of Luzon, Philippines, https://doi.org/10.21203/rs.3.rs-3827034/v1, 2024.
[89] G. James, D. Witten, T. Hastie, R. Tibshirani, An Introduction to Statistical Learning, vol. 112, springer, New York, 2013, p. 18.
[90] T. Hastie, R. Tibshirani, J.H. Friedman, J.H. Friedman, The Elements of Statistical Learning: Data Mining, Inference, and Prediction, vol. 2, springer, New York,
2009, pp. 1–758.
[91] Y. Freund, R.E. Schapire, Experiments with a new boosting algorithm, icml 96 (1996, July) 148–156.
[92] B. Charbuty, A. Abdulazeez, Classification based on decision tree algorithm for machine learning, Journal of Applied Science and Technology Trends 2 (1)
(2021) 20–28.
[93] C.M. Bishop, N.M. Nasrabadi, Pattern Recognition and Machine Learning vol. 4, springer, New York, 2006, p. 738.
[94] J.H. Friedman, Greedy function approximation: a gradient boosting machine, Ann. Stat. (2001) 1189–1232.
[95] T. Chen, C. Guestrin, Xgboost: a scalable tree boosting system, in: Proceedings of the 22nd Acm Sigkdd International Conference on Knowledge Discovery and
Data Mining (Pp. 785-794).Xgboost: A Scalable Tree Boosting System. In Proceedings of the 22nd Acm Sigkdd International Conference on Knowledge
Discovery and Data Mining, 2016, August, pp. 785–794.
[96] T. Chen, T. He, M. Benesty, V. Khotilovich, Y. Tang, H. Cho, T. Zhou, Xgboost: extreme gradient boosting, R package version 0.4-2 1 (4) (2015) 1–4.
[97] S. Ramraj, N. Uzir, R. Sunil, S. Banerjee, Experimenting XGBoost algorithm for prediction and classification of different datasets, Int. J. Control Theor. Appl. 9
(40) (2016) 651–662.
[98] P. Geurts, D. Ernst, L. Wehenkel, Extremely randomized trees, Mach. Learn. 63 (2006) 3–42.
[99] K. Torkkola, Linear discriminant analysis in document classification, in: IEEE ICDM Workshop on Text Mining, vol. 29, 2001, November.
[100] A. Patle, D.S. Chouhan, SVM kernel functions for classification, in: 2013 International Conference on Advances in Technology and Engineering (ICATE), IEEE,
2013, January, pp. 1–9.
[101] M. Schonlau, R.Y. Zou, The random forest algorithm for statistical learning, STATA J. 20 (1) (2020) 3–29.
[102] W. Xie, W. Nie, P. Saffari, L.F. Robledo, P.Y. Descote, W. Jian, Landslide hazard assessment based on Bayesian optimization–support vector machine in
Nanping City, China, Nat. Hazards 109 (1) (2021) 931–948.
[103] W. Sun, P. Bocchini, B.D. Davison, Applications of artificial intelligence for disaster management, Nat. Hazards 103 (3) (2020) 2631–2689.
[104] J. Cervantes, F. Garcia-Lamont, L. Rodríguez-Mazahua, A. Lopez, A comprehensive survey on support vector machine classification: applications, challenges
and trends, Neurocomputing 408 (2020) 189–215.
[105] M. Awad, R. Khanna, M. Awad, R. Khanna, Support vector machines for classification, in: Efficient Learning Machines: Theories, Concepts, and Applications for
Engineers and System Designers, 2015, pp. 39–66.
[106] Š. Raudys, How good are support vector machines? Neural Network. 13 (1) (2000) 17–19.
[107] E. Skoufias, Y. Kawasoe, E. Strobl, P. Acosta, Identifying the vulnerable to poverty from natural disasters: the case of typhoons in the Philippines, Economics of
Disasters and Climate Change 4 (2020) 45–82.
[108] J.P. Lapidez, J. Tablazon, L. Dasallas, L.A. Gonzalo, K.M. Cabacaba, M.M.A. Ramos, V. Malano, Identification of storm surge vulnerable areas in the Philippines
through the simulation of Typhoon Haiyan-induced storm surge levels over historical storm tracks, Nat. Hazards Earth Syst. Sci. 15 (7) (2015) 1473–1481.
[109] J.G.J. Molina, People’s knowledge, people’s defense: utilizing local practices for disaster safety and adaptation in Rapu-Rapu, Albay, Philippines, Jurnal Kajian
Wilayah 3 (2) (2016) 225–237.

24
E.A. Onsay and J.F. Rabajante International Journal of Disaster Risk Reduction 112 (2024) 104809

[110] W. Liu, S. Dugar, I. McCallum, G. Thapa, L. See, P. Khadka, P. Shakya, Integrated participatory and collaborative risk mapping for enhancing disaster
resilience, ISPRS Int. J. Geo-Inf. 7 (2) (2018) 68.
[111] G. Soriano, Disaster risk reduction knowledge among local people in a selected community in the Philippines, Journal of Health and Caring Sciences 1 (2)
(2019) 92–99.
[112] C.C. Labayo, Living in danger: exploring the culture of disaster of the ati peoples in Bicol, Philippines, Journal of International Development and Cooperation
25 (1) (2019) 1–12.
[113] L.B. Grefalda, J.M. Pulhin, E.P. Santos, Resilience to Weather-Related Disasters of a CBFM Community in Ligao, Albay, Philippines, 2018.
[114] V. Bollettino, T. Alcayna, K. Enriquez, P. Vinck, Perceptions of Disaster Resilience and Preparedness in the Philippines, Harvard Humanitarian Initiative,
Cambridge, MA, USA, 2018.
[115] C.B.A. Altavas, K.A.M. Lopez-Vito, J.A.S. Mendoza, S.R. Sarmenta Jr, The typhoon preparedness communication program of Albay province: a case study, in:
24th Annual Conference of the Asian Media Conference, Dubai, United Arab Emirates, 2015.
[116] K.E. Evasco, B.R. Alejandro IV, Disaster management in the high risk barangays in Albay Province, Philippines, ASIAN RURAL SOCIOLOGY IV (2010) 413.
[117] J.C.P. Espia, P. Fernandez, Insiders and outsiders: local government and NGO engagement in disaster response in Guimaras, Philippines, Disasters 39 (1)
(2015) 51–68.
[118] E.R. Florano, J.M.S. Perez, A.T.K. Pinheiro, Building back a better nation: disaster rehabilitation and recovery in the Philippines, in: Asian Law in Disasters,
Routledge, 2016, pp. 131–160.
[119] E.M. Preña, C.P. Labayo, Impacts of a ‘double disaster’on small-scale grouper farmers in the Bicol region, Philippines, Philippine Social Science Journal 5 (4)
(2022) 55–67.

25