H. L. Gururaj et al., Int. J. Transp. Dev. Integr., Vol. 6, No.

4 (2022), 363–377

PREDICTING TRAFFIC ACCIDENTS AND THEIR INJURY

SEVERITIES USING MACHINE LEARNING TECHNIQUES
H. L. Gururaj1, v. janhavi2, U. tanuja2, fransesco flamini3, B. C. soundarya2,
vinayAkumar ravi4
1
Department of Information Technology, Manipal Institute of Technology Bengaluru, Manipal Academy
of Higher Education, India
2
Vidyavardhaka College of Engineering, India
3
IDSIA USI-SUPSI, University of Applied Sciences and Arts of Southern Switzerland, Switzerland
4
Center for Artificial Intelligence, Prince Mohammad Bin Fahd University, Saudi Arabia

Abstract
Traffic accidents are among the most censorious issues confronting the world as they cause numerous
deaths, wounds and fatalities just as monetary misfortunes consistently. According to the world health
organization (WHO) reports, 5,18,3626 accidents took place in India in the year 2019. Factors that
contribute to these road crashes/ traffic accidents and resulting injuries include inattentive drivers, unen-
forced traffic laws, poor road infrastructure, driving in bad weather conditions and others. This investiga-
tion effort establishes models to select a set of influential factors and to build up a model for classifying
the severity of injuries. Machine learning models can be applied to model and predict the severity of
injury that occurs during road accidents. One such way is to apply unsupervised learning models such as
Apriori, Apriori TID (transaction id), SFIT (set operation for frequent itemset using transaction database)
and ECLAT (equivalence class clustering and bottom-up lattice traversal) which analyze the unlabeled
traffic accidents dataset and determine the relationship between traffic accidents and injury. This research
work is helpful for traffic departments to decrease the number of accidents and to distinguish the injury’s
seriousness extensive simulations were carried out to demonstrate the unsupervised learning algorithms
for predicting the injury severity of traffic accidents. Apriori algorithm predicts the patterns in 962 milli-
seconds, Apriori TID (transaction id) algorithm predicts the pattern in 557 milliseconds, SFIT algorithm
predicts the pattern in 516 milliseconds and ECLAT algorithm predicts the pattern in 124 milliseconds.
ECLAT algorithm took less time compared to all the other algorithms.
Keywords: Apriori, ECLAT, machine learning, traffic accidents

1 Introduction
Traffic accidents cause considerable economic, monetary and social losses to individuals,
families and the nation as a whole. As indicated by the death approximation provided by
the world health organization (WHO), there is an alarming increase in the number of traffic
accidents [1][2]. Each year the lives of approximately 1.2 million are lost with 50 million
individuals injured [1].
Road traffic accidents have reached alarming levels every year across the world. Every year,
roughly 1.2 million people are dies due to road traffic crashes worldwide. A similar trend can
be seen in countries with the most traffic accident-related death rate per 100,000 namely Zim-
bambwe (61.90), Liberia (52.03), Malawi (51.62), Gambia (47.51), Togo (46.62), Tanzania
(46.17), Rwanda (45.90), Sao tome (45.52), Burkina Faso (44.94) and Burundi (44.94) [3].
The public is suffering from many major injuries even after many years of the accident [4].
Consequently, road crashes became the main source of human death and injuries all around
the world.
There are several factors due to which accidents occur. These are: rear-end collisions, side-
impact collisions, side swap collisions, rollover, hit and run, head-on collisions, single-car
accidents, multiple vehicle pile-ups and drunk and drive. There are two types of accidents
namely: Major accident injury and minor accident injury. Major accident injury is one in

© 2022 WIT Press, www.witpress.com

ISSN: 2058-8305 (paper format), ISSN: 2058-8313 (online), http://www.witpress.com/journals
DOI: 10.2495/TDI-V6-N4-363-377
364 H. L. Gururaj et al., Int. J. Transp. Dev. Integr., Vol. 6, No. 4 (2022)

which at least one casualty endures genuine injury requiring hospitalization. Minor accident
injury is one in which the victim(s) doesn’t require hospitalization.
Figure 1 represents the graph of traffic accidents categorized by three parameters i.e., road
accidents, the person dies and injuries that happened on roadways from 2014 to 2019, the
data provided by the Government of India (ministry of road transport and highways research
wing, New Delhi). The descriptive model was developed using clustering techniques and
association learning techniques [5][6]. Around 140 lives have been lost across the country
due to road crashes. Different states which recorded maximum traffic accidents are Delhi,
Maharashtra, Gujarat, Assam, Kerala, Karnataka, Rajasthan, Punjab and Tamil Nadu [9][10].
Figure 2 represents the different traffic accidents with related injuries. For example, being
drunk and driving may lead to some fractures or may lead to death. Different types of injuries
are brain injuries, spine fracture, pelvic fracture, back and spinal cord trauma, skull and max-
illofacial, rib fracture, broken bones, whiplash, scrapes and cuts, internal bleeding, and herni-
ated disc, knee trauma, soft tissue injuries, etc. The existing system used some of the machine
learning algorithms like the random forest (RF), K-nearest neighbor and decision tree to
identify the accidents, accidents vehicles, and injuries which provides less accuracy [7][8].
The proposed work is a real-time application that is helpful for the government sector to
decrease the number of traffic accidents and its injuries severity. Unsupervised learning is
a machine learning technique, which helps to discover the hidden patterns and information
in the given datasets. Mainly unsupervised learning algorithm needs only data sets without
trained or labeled and tested which does not consume time for calculating, analyzing and pre-
dicting accidents severities. We use association rule mining which predicts the severities of
road crashes [11][14]. Association rule mining is used to discover the relationships between
data items within large datasets. Association rule uses support and confidence to identify the
most important relationships of data items in less amount of time. The main idea of this work

Figure 1: Trends of traffic accidents, the person dies, and injuries by road category from 2014
to 2019.
 H. L. Gururaj et al., Int. J. Transp. Dev. Integr., Vol. 6, No. 4 (2022) 365

Figure 2: Different traffic accidents with injuries.

is to minimize traffic accidents and their injuries because the public is suffering from many
major injuries even after many years of accidents [17][18]. We took four machine learning
algorithms i.e., Apriori, Apriori TID (transaction id), SFIT (set operation for frequent itemset
using transaction database) and ECLAT (equivalence class clustering and bottom-up lattice
traversal) which analyze the unlabeled traffic accidents dataset and determine the relationship
between traffic accidents and injury.
This finding of the paper is as follows:

• Identify the correlation between the accident types i.e., drunk and drive may lead to hit
and run, etc.
• Discovering the relationship between the different categories of injuries i.e., brain injury
may lead to spinal cord injury, etc.
• Identifying the correlation between accident types with injuries i.e., collision may lead to
brain injury, broken bones, etc.

Our work can be helpful in several future scenarios to save people’s lives by early prediction
of the type of accidents with related injuries to get proper treatment because so many people
are suffering from major injuries even after many years of the accident. The paper is organ-
ized as follows: section 2 represents related work, section 3 describes proposed work, section
4 presents a comparative analysis of existing methodologies, section 5 shows the results and
section 6 presents a comparative analysis of proposed methodologies. Finally, we concluded
this paper.

2 Related work
In this section, we briefly discussed existing methodologies and their contributions. We also
acknowledge the limitations of existing methodologies.
Xiao Li, et.al, [11] identify the impact of traffic accidents and injury severities which is
dependent on three angles, example, daily travelers (included age, sex, and so forth.), vehi-
cles (included vehicle type and number, vehicle transmission and so on.) and street (included
pavement condition, cross-area type and so forth.).
366 H. L. Gururaj et al., Int. J. Transp. Dev. Integr., Vol. 6, No. 4 (2022)

Siddharth Tripathi, et.al, [12] proposed a model called CBIT (cloud-based intelli-
gent traffic system) to detect traffic accidents and alerts the traffic authorities of the location
of an accident. They took three principle targets: examining emission, accident detection and
unique vehicle ID. The proposed model used some sensors i.e., NOX, SOX, CO and tempera-
ture sensors like MQ135, MQ3 and LM35 for detecting the emission. Accelerometer (ADC
MCP3008) and force resistive sensors are used to detect accidents. Unique vehicle ID will
be assigned to all the vehicles which are helpful for traffic authorities to monitor the vehicle.
GPS (UBLOX G7020) and GSM (SIM 900) modules are used for sharing the location of
accidents with traffic authorities. The WiFi port of raspberry pi is used to aggregate the infor-
mation and then it is stored in the cloud.
Naji Taaib Said Al Wadhahi, et.al, [13] proposed a model for the detection of accidents and
prevention system to lessen traffic hazards utilizing Arduino boards (Atemga 328) and some
sensors such as IR. This framework includes two stages: detection and prevention of traffic
accidents. Here is a sensor (IR) that helps to identify the accidents and alarm the individuals
by sending messages. Messages are sent through GSM (SIM 900D) module. These sensors
notify the driver about accidents.
JIAN ZHANG, et.al, [14] compared injuries severities predictions using statistical meth-
ods and machine learning algorithms. The accident seriousness, street geometry and traffic
status were gathered at road separate territories in Florida [24]. They assessed the two most
utilized statistical methods which are the multinomial logit model and the ordered probit
(OP) model. They have used machine learning algorithms i.e., K nearest neighbor (KNN),
decision tree, RF and support vector machine (SVM) to predict the accident injury severities.
The RF method is best for large and extreme accidents while the OP model was the most
fragile one.
Guang Yu, et.al, [15] proposed a strategy for taking gander blockage and controlling traffic
accidents. Speed and lane changing control systems are used to reduce traffic accidents. They
took a portion of the countermeasures to keep away from the auto collisions through admin-
istration of path, controlling rate of crashes, adjusting of speed and traffic data.
Fabio Galatioto, et.al, [16] proposed a model i.e., MAIA (model and methods for accident
prediction and its impact assessments) toolkit for predicting traffic accidents and their injury
severities. The authors in this paper concentrated on three different problems: accident rate
prediction, estimation of both injury and non-injury collisions and estimation of collision
security level. Finally, we discussed existing works and their contributions. We surveyed so
many papers and their solutions which are related to our work. There are so many issues in
existing work, so we compare these solutions with our proposed solutions. Our work has
made more contributions to these issues.

3 Proposed work
Road traffic accidents and their injuries are the leading cause of death. The proposed research
work is a real-time application that is helpful for traffic departments to decrease the number of
traffic accidents and distinguishes the injury severities. This application is also helpful for doc-
tors to give better treatment. The proposed system makes use of association learning methods to
discover the hidden patterns between traffic accidents and related injury severities.
Figure 3 shows the design for anticipating traffic injury severities. The system predicts
the traffic accidents and their injury severities dependent based on old datasets. The users
here will be the Admin, Member (City Traffic In Charger, Doctors) and Visitor (Public). The
administrator maintains the entire system. The administrator is responsible for the city in
charge of creating and uploading the necessary data for processing. Member (City Traffic In
 H. L. Gururaj et al., Int. J. Transp. Dev. Integr., Vol. 6, No. 4 (2022) 367

Figure 3: Architecture diagram of traffic accident severity.

Charger, Doctors) City In Charge is a service receiver and responsible for uploading the traf-
fic accidents data into the server. Visitor (Public) Visitor is a user who visits the application.
The visitor has limited accessibility.

3.1 Methodology

Unsupervised learning methods help to extract the relevant information and structure in unla-
beled datasets. It takes less time to extract the patterns from the given datasets. Association
rule discovers the patterns using support and confidence [19]. In this paper, we are using four
types of algorithms i.e., Apriori [20], Apriori TID (transaction id) [21], SFIT (set operation
for frequent itemset using transaction database) [22,29] and ECLAT (equivalence class clus-
tering and bottom-up lattice traversal) [26] to discover the patterns of traffic accidents and its
injury severities.

3.1.1 Apriori algorithm

Apriori algorithm is one of the association rule learning [20]. Defining association rule
mining is as follows:
Let I item = {i1, i2, i3,…….,in}set of n attributes (items). Let D transactions = {t1, t2, t3,…….,tn}
set of n transactions (database). Different attributes are used in the Apriori algorithm i.e.,
Itemset, K-itemset and null sets where the collection of more items including zero is called
itemsets. Itemsets include k-items is called k-itemsets. Itemset that does not include any
items called null (empty) set.
In each transaction T contains a subset of items taken from I. Transaction T includes an
item set A. If A is a subset of T then the support count σ (A) can be written mathematically,

σ ( A) = |{ti |A ⊆ ti , ti ∈ T } | (1)

Where | | indicated the number of items in a set. Transactions of items in an itemset with
support [32] [33].

S(A) = σ (A) / N

Where S(A) is greater than or equal to some defined threshold.

368 H. L. Gururaj et al., Int. J. Transp. Dev. Integr., Vol. 6, No. 4 (2022)

An association rule for the Apriori algorithm with an expression form A->B where A and
B are disjoint itemsets i.e., A ∩ B = ø. It can be measured by using support and confidence.
The definition of support and confidence are:

σ ( A ∪ B)
Support s(A -> B) = (2)
N

σ ( A ∪ B)
Confidence c(A -> B) = (3)
σ ( A)

In the above equation (i) and (ii) where support helps to measure the rule in a given itemset
and confidence discovers frequently appearing items in a transaction T.
Figure 4 shows the processing of the Apriori algorithm. The data processing means extract-
ing relevant data from the accident database or server. Now applying the Apriori algorithm for
generating rules to predict the accident patterns and represents the results of generated patterns.

Algorithm 3.1 Apriori Algorithm for generating frequent itemsets

STEP 1: Examine old data sets then regulate the support (s) for each item.
STEP 2: Create L1 (only one item set).
STEP 3: First use Lk-1, and then unite Lk-1 to create candidate K sets.
STEP 4: Examine the generated candidate K item set and then create support for the can-
didate in each item set K.
STEP 5: Append item set frequently, till C = Null set (ø).
STEP 6: Create all non–empty subsets for frequent itemsets.
STEP 7: Examine the confidence for all non-empty subsets. If the examined confidence is
higher or equal to the given confidence, then append to the association rule.

3.1.2 Apriori TID Algorithm

Apriori algorithm discovers all the patterns or relations in the given item set. It also discovers
the number of candidate items [21].
Suppose Ttransactions {t1, t2, t3,…..tk }, (K ≥1) ( transaction sets, Titem (Ti) = (I1, I2, I3, …., Im),
(m ≥ 1) itemsets and K-itemsets (In) = {i1, i2, i3,…..in }, (n ≥ 1) k-itemsets where Kitemsets ⊆ I.
Figure 5 shows the processing of an Apriori TID (transaction id) algorithm. Examine the
old data items to regulate the support(s) for each item. Find C1, C2, C3,…..Ck and then

Figure 4: Processing of Apriori algorithm.

 H. L. Gururaj et al., Int. J. Transp. Dev. Integr., Vol. 6, No. 4 (2022) 369

Figure 5: Processing of Apriori TID (transaction id) algorithm.

generate L1, L2, L3,…..Lk-1 for frequent item set. Generate candidate Ck set by using Lk-1.
Append item set frequently till C = Null set (ø). If Ck ≥ min-support then append to the asso-
ciation rule and displays the results of generated patterns [30][31].

Algorithm 3.2 Apriori TID Algorithm for generating frequent itemsets

STEP 1: Examine old data sets then regulate the support (s) for each item.
STEP 2: Find C1’
STEP 3: Generate L1 (Frequent item set).
STEP 4: Find C2’, C3’ …..
STEP 5: First use Lk-1, and then unite Lk-1 to create candidate K sets.
STEP 6: Examine the generated candidate K item set and then create support for the candi-
date in each item set K, by comparing with the previous step (but not with the original data-
set as we did in the Apriori algorithm).
STEP 7: Append item set frequently, till C = Null set (ø).
STEP 8: Create all non–empty subsets for frequent itemsets.
STEP 9: Examine the confidence for all non-empty subsets. If the examined confidence is
higher or equal to the given confidence, then append to the association rule.

3.1.3 SFIT (Set operation for frequent itemset using transaction database)
SFIT algorithm is a combination of apriori algorithm which is used for data mining and dif-
ferent set operations like union and intersection are used [22].
For constructing K-itemsets, we use frequent itemsets (K-1). Union is formed for K and
(K-1) itemsets. Employing intersection operation for transaction identifiers (tids) of itemsets.
Itemsets {M} transactions with Tid 2, 4, 6, 8 and {N} transactions with Tid 1, 2, 3, 5, 6,
7, 8 i.e., T(M) = {2, 4, 6, 8} and T(N) = {1, 2, 3, 4, 5, 6, 7, 8}. The itemsets {M, N} is the
union of MN itemsets. In order to find out the tids for {M, N} by using intersection principle
as shown below:

T(MN) = T(M) ∩ T(N)

= {2, 4, 6, 8} ∩ {1, 2, 3, 4, 5, 6, 7, 8}
370 H. L. Gururaj et al., Int. J. Transp. Dev. Integr., Vol. 6, No. 4 (2022)

T(MN) = {2, 4, 6, 8}

If the obtained output is greater than the given minimum support, then it will be included in
the frequent itemset otherwise it will be eliminated [29].

Algorithm 3.3 SFIT Algorithm for generating frequent itemsets

STEP 1: Examine old data sets then regulate the support (s) for each item.
STEP 2: Find C1 (candidate item).
STEP 3: Generate L1 (based on C1 finding L1 i.e., frequent one item set).
STEP 4: Find C2, C3…….., (based on C2……..Cn finding L2…..Ln).
STEP 5: Find C4, when C4 = ø (null set) then stop the iteration.
STEP 6: Find frequent itemset (L) and subsets showing the relationship between items.
STEP 7: Finally compare with minimum specified confidence, the rules where confidence
is greater than or equal, add to the strong association rule.

3.1.4 ECLAT (Equivalence class clustering and bottom-up lattice traversal)

ECLAT algorithm is an acronym for equivalence class clustering and bottom-up lattice tra-
versal [24][25]. In ECLAT algorithm, it will make use of vertical transactions id (tid) sets inside
the database, clustering of equivalence classes and lattice traversal (bottom-up approach). This
algorithm reduces storage and cost [26]. ECLAT algorithm metamorphose the parallel data-
base into perpendicular database i.e., from item set format Apriori<TIDx, X1, X2,…..,Xk> to
transaction id(tid) set format ECLAT<Xk, TID1, TID2,……,TIDk>. ECLAT algorithm uses a
vertical approach tidset database includes a list of items: the set of all transaction identifiers
called tidset of A, which is represented as tidset(A) = {Ti, Tid | Ti ∈ Di X ⊆ Ti}. Several ele-
ments in tidset (A) are called support of A i.e., represented as σ (A) = |tidset (A)|.
Figure 6 shows the processing of an ECLAT algorithm. From the accident database or
server extracting the relevant information from the database is called data processing then
applying the éclat algorithm for generating association rules that predict the accident patterns
and represent the results of generated accident patterns.
Algorithm 3.4 ECLAT Algorithm for generating frequent itemsets
Step 1: Generate a tidlist for every item by scanning the database.
Step 2: Tidlist of {a} is precisely the item set of transactions {a}.
Step 3: Bisect the tidlist of {a} with the other items of tidlist, the resulting of dividing tidlist
is {a,b}, {a,d}, {a,c}…..

Figure 6: Processing of ECLAT algorithm.

 H. L. Gururaj et al., Int. J. Transp. Dev. Integr., Vol. 6, No. 4 (2022) 371

Step 4: Recurrent 1 on {a} – given in the database.

Step 5: Replicate these steps for all other item sets.

4 comparative analysis of the existing methodologies

In this section, Table 1 shows the performance of existing methodologies.

Table 1: Comparative analysis of other methodologies.

No. Data sets used Algorithms or Accuracy Limitations

techniques used
[11] >300 datasets of KNN, decision KNN – 80.26%
Small Dataset,
traffic accidents. tree and Decision tree
hence, less
association rule. – 73.68% accurate results.
[12] 500 traffic Raspberry pi, 70.1% The cost of
accident data. emission sensors, hardware
GPS and GSM components is
modules. very high.
[13] 5,000 datasets of KNN, decision Overall accuracy Used labeled/
traffic accidents tree, RF, SVM. ranges from trained datasets
44.7% to 80.5% that lead to very
low accuracy.
[14] >6,000 datasets of RF, logistic Logistic regression This model
Michigan traffic regression, naïve – 74.5% is using only
accidents. bayes and ada Naïve Bayes Michigan datasets
boost. – 73% and has less
Ada boost accuracy.
– 74.5%
RF – 75%
[15] 3,643 traffic Bayesian network Overall accuracy Using fewer china
accident data of and information ranges from 50 datasets leads to
china. entropy. to 90% less accuracy.
[16] 1,130 traffic Convolution 87% Used fewer
accident datasets. neural network amounts of datasets
(CNN). which leads to low
accuracy.
[17] 7,000 traffic Feedforward neural FNN – 69% They used both
crashes datasets of network (FNN), FNN-FCM training and
the UK SVM, fuzzy c – 70.50% testing datasets
means clustering- SVM – 73% but the accuracy
based feed-forward SVM-FCM – 74% is less compared
neural network to the proposed
(FNN-FCM) and methodology.
fuzzy c means
based support
vector machine
(SVM-FCM).
372 H. L. Gururaj et al., Int. J. Transp. Dev. Integr., Vol. 6, No. 4 (2022)

Table 1: (Continued)

No. Data sets used Algorithms or Accuracy Limitations

techniques used
[18] 150 traffic Convolution 78.5% Even though the
accident trained neural network datasets are small,
datasets. (CNN). this method gives
less accuracy.
[19] >10,000 traffic Naïve bayes, RF, NB – 74% This framework
accident datasets MLP, Ada boost. RF – 77% predicts traffic
of USA. MLP – 77% accident severity
AB – 75% in the USA but the
accuracy is less.

[20] > 4,000 traffic Ordered a probit OP – 80% Accuracy is

accident datasets. model, an artificial ANN – 86.50% less compared
neural network EM – 87% to the proposed
(ANN) and an methodology.
ensemble model.

5 results analysis
In this paper, all the below figures represents the outcome of our proposed methodology. For
predicting the injury patterns of traffic accidents, we are using unsupervised learning or asso-
ciation mining rules. The unsupervised learning method gives relevant information from the
given datasets without training and testing those datasets. We took four algorithms to predict
the traffic accident severities. All the four algorithms Apriori, Apriori TID, SFIT and ECLAT
generate the patterns in three ways i.e., firstly, accident type with injuries for example colli-
sion may lead to spine fracture with 75% confidence means that 75% possibility of people
frequently met with this accident severity. Second, discovering the relationship between the
different categories of injuries for example brain injury may lead to spinal cord injury with
85% of confidence. The third category is to find out the relation between the accident types
for example drunk and driving may lead to hit and run with 90% of confidence. Almost all the
algorithms will generate the same kind of patterns but the difference is efficiency will vary.
In this paper, we find out that the éclat algorithm is the best mining method to generate the
patterns because it takes less storage capacity, cost and time.
Figure 7 shows the prediction of traffic accident severity patterns using ECLAT algorithm.
This algorithm uses bottom up approach for mining frequent itemsets. ECLAT algorithm
does not scan whole database to generate support values. The output of this ECLAT algo-
rithm showed in three ways i.e., accident type with injuries (collision may leads to brain inju-
ries and broken bones with confidence 100%), discovering the relation between different type
of injuries (brain injury may leads to broken bones with confidence 100%) and also it helps to
find out the accident types (drunk and drive may leads to hit and run with 100% confidence).
It is best suitable for large and small datasets and discovers the pattern in 124 milliseconds
 H. L. Gururaj et al., Int. J. Transp. Dev. Integr., Vol. 6, No. 4 (2022) 373

as shown in above figure. This algorithm is best for mining when compare to Apriori, Apriori
TID and SFIT algorithm because it takes less storage capacity, cost and less time.
As per the survey of the World health organization (WHO), Figs 8 and 9 shows the causes
of traffic accidents. Different types of accidents due to collisions, violations of traffic rules
and highways [28]. These are some of the old datasets which are taken from the government
survey document to show how the traffic accidents took place and also shows the number of
accidents, the person dies and the person injured from 2017 to 2018.

Figure 7: Predicting severity patterns using the ECLAT algorithm.

Figure 8: Different types of traffic accident collision of 2017 and 2018.

374 H. L. Gururaj et al., Int. J. Transp. Dev. Integr., Vol. 6, No. 4 (2022)

Figure 9: Traffic accidents due to traffic rules violations during 2017 and 2018.

Figure 10: Time comparison between Apriori, Apriori TID, SFIT and ECLAT algorithm.

6 comparative analysis of proposed method

In this paper, we have done various experiments to scrutinize the performance of the pro-
posed algorithms. The number of datasets 100, 1,000, 4,000 and 10,000 were taken to com-
pare all the algorithms (Apriori, Apriori TID, SFIT and ECLAT) in terms of execution time
represented in different colors.
Figure 10 shows the graphical representation of the execution time of all four algorithms.
ECLAT algorithm is the best for finding out the frequent itemsets compared to other algo-
rithms because the éclat algorithm took less time to discover the patterns of traffic accident
severities from the given accident datasets.
 H. L. Gururaj et al., Int. J. Transp. Dev. Integr., Vol. 6, No. 4 (2022) 375

7 conclusion
Road safety represents a significant part of our lives, so it is necessary to reduce traffic acci-
dents and their injury severities. The proposed work is a real-time application that is helpful
for traffic departments as well as doctors to reduce traffic accidents and for providing proper
treatments for injuries. In this paper, we used unsupervised learning methods to discover the
patterns of traffic accidents and their injuries from the given datasets. The simulation results
of the proposed methods for predicting traffic accident severities i.e., the Apriori algorithm
predicts the patterns in 962 milliseconds, the Apriori TID algorithm predicts the pattern in
557 milliseconds, the SFIT algorithm predicts the pattern in 516 milliseconds and the ECLAT
algorithm in predicts the pattern in 124 milliseconds. ÉCLAT algorithm is the best algorithm
for discovering patterns of a traffic accident and its injury severities from the given traffic
accident datasets, because it takes less time, cost and storage.
The proposed system in the future can be enhanced with a module like the Public notifica-
tion module, Query Module and Predicting reasons for accidents. Public notification (SMS/
Email): We can add a public notification that helps people who met with an accident and
they can get help to reach the hospital and get treatment. Query module: We can add a query
module for the interaction between administrator and member (Traffic in charger or Doctor)
to maintain more traffic accident records and the public can get better treatment for a particu-
lar predicted injury. We can also predict reasons for accidents which helps traffic departments
to take precautionary measures.

Acknowledgements
No financial support was received to perform the research work of this manuscript.

REFERENCES
[1] Zhou, Z, Attention based stack resnet for citywide traffic accident prediction. 20th IEEE
International Conference on Mobile Data Management (MDM), pp. 369-370, 2019.
[2] Amundsen, Finn Harald, and Guro Ranes, Studies on traffic accidents in Norwegian
road tunnels. Tunneling and Underground Space Technology, pp. 3-11, 2000.
[3] Myounghoe Kim, Seongwon Lee, Jaehyun Lim, Jongeun Choi, and Seong Gu
Kang, Unexpected Collision Avoidance Driving Strategy Using Deep Reinforcement
Learning. IEEE Access, pp. 17243–17252, 2020.
[4] Wang, H., Huang, Y., Khajepour, A., Rasekhipour, Y., Zhang, Y. & Cao, D., Crash
Mitigation in Motion Planning for Autonomous Vehicles. IEEE Transactions on
Intelligent Transportation Systems, pp. 1–11. 2020.
[5] Albatayneh, O., Moomen, M., Farid, A., & Ksaibati, K., Complementary Modeling
of Gravel Road Traffic-Generated Dust Levels Using Bayesian Regularization
Feedforward Neural Networks and Binary Probit Regression. International Journal of
Pavement Research and Technology, 6, pp. 1-8, 2018.
[6] Fernandes, Fábio & Alves de Sousa, Ricardo & Ptak, Mariusz., Head Injury Simulation
in Road Traffic Accidents. Springer, pp. 59-48, 2018.
[7] Li, G., Ma, H., Guan, T. et al., Predicting Safer Vehicle Font-End Shapes for Pedestrian
Lower Limb Protection via a Numerical Optimization Framework. Int.J Automota
Technol, 21, pp. 749–756, 2020.
[8] A., Fang, C., Qiu, L., Zhao and Y., Jin, Driver Risk Assessment Using Traffic Violation
and Accident Data by Machine Learning Approaches. 3rd IEEE International
Conference on Intelligent Transportation Engineering (ICITE), pp. 291-295, 2018.
376 H. L. Gururaj et al., Int. J. Transp. Dev. Integr., Vol. 6, No. 4 (2022)

[9] Y., He, Z., Liu, X., Zhou and B., Zhong, Analysis of Urban Traffic Accidents Features and
Correlation with Traffic Congestion in Large-Scale Construction District. International
Conference on Smart Grid and Electrical Automation (ICSGEA), pp. 641-644, 2017.
[10] T., Jian, L., Zhi-qiang, X., Jian-feng and G., Hong-yu, Cause Analysis and
Countermeasures of fatal Traffic Accidents on Road Passenger Transportation Based
on Typical Cases. 25th International Conference on Transportation Information and
Safety (ICTIS), pp. 951-955, 2019.
[11] Masoumi, K., Forouzan, A., Barzegari, H., Asgari Darian, A., Rahim, F., Zohrevandi,
B. & Nabi, S., Effective Factors in Severity of Traffic accident-related Traumas an
Epidemiologic Study Based on the Haddon Matrix. Emergency (Tehran, Iran), pp. 78–
82, 2018.
[12] Tripathi, S., Shetty, U., Hasnain, A. & Hallikar, R., Cloud-Based Intelligent Traffic
System to Implement Traffic Rules Violation Detection and Accident Detection Units.
3rd International Conference on Trends in Electronics and Informatics (ICOEI), pp.
22-26, 2019.
[13] Wadhahi, N.T.S.A., Hussain, S.M., Yosof, K.M., Hussain, S.A., & Singh, A.V.,Accidents
Detection and Prevention System to reduce Traffic Hazards using IR Sensors. 2018
7th International Conference on Reliability, Infocom Technologies and Optimization
(Trends and Future Directions) (ICRITO), pp. 254-257, 2018.
[14] Zhang, J., Li, Z., Pu, Z., & Xu, C., Comparing Prediction Performance for Crash Injury
Severity among Various Machine Learning and Statistical Methods. IEEE Access, pp.
1–1, 2018.
[15] Guang Yu, Shauo lu, Speed and lane changing control optimization research on accident
and congestion concentrated section of guang-shen freeway, IEEE Access, pp. 1-8,
2019.
[16] Galatioto, F., Catalano, M., Shaikh, N., McCormick, E., & Johnston, R., Advanced
accident prediction models, and impacts assessment. IET Intelligent Transport Systems,
12(9), pp. 1131–1141, 2018.
[17] M., Zheng, et al., Traffic Accident’s Severity Prediction: A Deep-Learning Approach-
Based CNN Network. IEEE Access, pp. 39897-39910, 2019.
[18] Leijdesdorff, H. A., Gillissen, S., Schipper, I. B., & Krijnen, P., Injury Pattern and Injury
Severity of In-Hospital Deceased Road Traffic Accident Victims in The Netherlands:
Dutch Road Traffic Accidents Fatalities. World Journal of Surgery, 6, pp. 268-272,
2020.
[19] Zong, F., Chen, X., Tang, J., Yu, P. & Wu, T., Analyzing Traffic Crash Severity With
Combination of Information Entropy and Bayesian Network. IEEE Access, pp. 63288–
63302, 2019.
[20] Gao, Z., Pan, R., Yu, R. & Wang, X., Research on Automated Modeling Algorithm
Using Association Rules for Traffic Accidents. IEEE International Conference on Big
Data and Smart Computing (BigComp), pp. 1-6, 2019.
[21] Song, J., Wen, R. & Yan, W., Identification of Traffic Accident Clusters using Kulldorff’s
Space-Time Scan Statistics. 2018 IEEE International Conference on Big Data (Big
Data), pp. 1-6, 2018.
[22] Tai, W.-K., Wang, H.-C., Chiang, C.-Y., Chien, C.-Y., Lai, K. & Huang, T.-C., RTAIS:
Road Traffic Accident Information System. 2018 IEEE 20th International Conference
on High-Performance Computing and Communications, pp. 1-8, 2017.
 H. L. Gururaj et al., Int. J. Transp. Dev. Integr., Vol. 6, No. 4 (2022) 377

[23] Wenqi, L., Dongyu, L. & Menghua, Y., A model of traffic accident prediction based on
convolutional neural network. 2017 2nd IEEE International Conference on Intelligent
Transportation Engineering (ICITE), pp. 67-72, 2017.
[24] Xia, X.-L., Nan, B. & Xu, C., Real-Time Traffic Accident Severity Prediction Using
Data Mining Technologies. 2017 International Conference on Network and Information
Systems for Computers (ICNISC), pp. 1-8, 2017.
[25] Xu, Q., Zheng, J. & Sun, C., Digital modeling analysis of urban road traffic capacity
under the condition of traffic accidents. 2017 29th Chinese Control And Decision
Conference (CCDC), pp. 10-12, 2017.
[26] Singh, Sanjay, Road Traffic Accidents in India: Issues and Challenges. Transportation
Research Procedia. 25, pp. 4712-4723, 2017.
[27] Sanjana, K.R., S., Lavanya, and Y., Bevish Jinila, An approach on automated rescue
system with intelligent traffic lights for emergency services. 2015 International
Conference on Innovations in Information, Embedded, and Communication Systems
(ICIIECS), pp. 1-4, 2015.
[28] E., Abdullah and A., Emam, Traffic Accidents Analyzer Using Big Data. 2015
International Conference on Computational Science and Computational Intelligence
(CSCI), pp. 392-397, 2015.
[29] Alkandari, Abdulrahman Abdullah; Aljandal, Meshari, Theory of Dynamic Fuzzy
Logic Traffic Light Integrated System with Accident Detection and Action. IEEE
2015 Second International Conference on Computing Technology and Information
Management (ICCTIM) - Johor, pp. 62–68, 2015.
[30] Fu Huilin, Zhou Yucai, The Traffic Accident Prediction Based on Neural Network.
IEEE Access, pp. 1-6, 2011.
[31] Hui, Chen; Xianghui, Wang; Xiqiang, Zhang; Shaoli, Zhang, The Evaluation of Chinese
Urban Traffic Management System Application Based on Intelligent Traffic Control
Technology. 7th International Conference on Intelligent Computation Technology and
Automation, pp. 791–795. 2014.
[32] Li, Jikuang Yang, Otte, D., Zhou Xuenong, Zhang Chao, Yu Xiaoping, Li Weiqiang,
A Study on Motor-Scooter Accidents in China and Germany. IEEE 2013 Fifth
International Conference on Measuring Technology and Mechatronics Automation
(ICMTMA 2013), pp. 110–113, 2013.
[33] Sherif, Hossam M.; Shedid, M. Amer; Senbel, Samah A., Real-time traffic accident
detection system using a wireless sensor network. IEEE 2014 6th International
Conference of Soft Computing and Pattern Recognition (SoCPaR), pp. 59–64, 2014.
[34] Xie, Fang, Feng, Qingdong, Research of effects of the accident on traffic flow
characteristics. IEEE 2013 International Conference on Mechatronic Sciences, Electric
Engineering and Computer (MEC) - Shengyang, China Proceedings 2013 International
Conference on Mechatronic Sciences, pp. 1670–1674, 2013.