Universidad de Lima
Facultad de Ingeniería
Carrera de Ingeniería Industrial
A PRODUCT NETWORK ANALYSIS USING A
PRIORI ALGORITHM FOR EXTENDING
THE MARKET BASKET IN RETAIL
Tesis para optar el Título Profesional de Ingeniero Industrial
Gonzalo Manuel, Infante Caldas
Código 20180936
Xiomara Dayana, Molina Rubio
Código 20181210
Asesor
Yvan Jesús García López
Lima – Perú
Julio de 2023
Proceedings of the First Australian International Conference on Industrial Engineering and Operations
Management, Sydney, Australia, December 20-21, 2022
A Product Network Analysis Using A Priori Algorithm for
Extending the Market Basket in Retail
Gonzalo Manuel Infante Caldas
Universidad de Lima
Carrera de Ingeniería Industrial
Lima, Perú
[email protected] (https://orcid.org/0000-0002-5315-042X)
Xiomara Dayana Molina Rubio
Universidad de Lima
Carrera de Ingeniería Industrial
Lima, Perú
[email protected] (https://orcid.org/0000-0003-4524-0844)
Yvan Jesus Garcia Lopez
Universidad de Lima
Carrera de Ingeniería Industrial
Lima, Perú
[email protected] (https://orcid.org/0000-0001-9577-4188)
José Antonio Taquía Gutiérrez
Universidad de Lima
Instituto de Investigación Científica
Lima, Perú
[email protected] (https://orcid.org/0000-0002-1711-6603)
Abstract
Market basket analysis provides an insight into customer consumption patterns and trends in the industry. These will
be achieved by analyzing and studying the performance of the large datasets of transactions made by consumers held
in retail stores. These commercial transactions will be analyzed using the Machine Learning technique called the A
priori algorithm by establishing association rules and determining those groups of items in a market basket whose
association could represent better economic benefits for companies. This study will analyze the historical sales data
of the product groups, in order to identify relationships that al-low companies in the sector to generate patterns to
propose the increase of their portfolio based on the products with the greatest purchasing trends. At the end of this
investigation, commercial strategies will be proposed to improve sales, take advantage of spaces in stores and
implement more effective strategic offers, based on the groups of articles with the best associations found.
Keywords
Machine learning, basket analysis, portfolio extension, retail, a priori algorithm.
© IEOM Society International 2025
Proceedings of the First Australian International Conference on Industrial Engineering and Operations
Management, Sydney, Australia, December 20-21, 2022
1. Introduction
The retail industry is facing great changes as a result of the abnormal fluctuation in demand and the change in consumer
habits as a result of the COVID 19 pandemic. In addition, there is the constant struggle of companies in the sector to
maintain and increase their profits and offer the products and services that the customer wants to obtain (Kim, 2020).
This highlights the need for retailers to provide a more specialized service, offer a greater variety of products and
increase their sales are the main problems to be faced in this generation of post-pandemic organizations. Likewise, the
need to use data and sales analysis tools based on data regression algorithms is observed, these algorithms being part
of Machine Learning methodologies.
Nowadays, with the increase of globalization and the advance of technology, retail companies are constantly struggling
to maintain and increase their profits. Having large amounts of data collected from commercial transactions, gives rise
to the need to analyze them to extract considerations of products and consumer purchases, this can occur through the
application of Machine Learning techniques (Moldenhauer and Zwirnmann, 2019).
This research will seek to use the Machine Learning methodology in companies in the retail sector to forecast the
demand for certain products, analyze the historical sales data and its variation, in order to evaluate the products with
the greatest potential to expand the customer's market basket. Currently, there are many tools within the study
methodology, however, in this report the use of the A priori Algorithms tool will be evaluated. By performing the
analysis of the shopping basket and employing Apriori Algorithms to determine the associations of the different
products, the portfolio extension will proceed and will allow retail companies to know consumer behavior and plan
the sales and marketing strategies to be applied to have greater visibility and revenue (Moldenhauer and Zwirnmann,
2019).
1.1 Objectives
General objective: Analyze the correlation between increasing the product network and expanding the purchase
network within companies in the retail sector, in order to generate healthy margins in the industry and increase the
brands present in the cata-log of retail companies.
Specific objectives:
• Propose the increasement of the portfolio of new national and imported products in the study sector.
• Identify strategies that increase sales and generate healthy margins in companies in the retail sector.
• Interpret and compare the main statistical indicators of the A priori Algorithms
2. Literature Review
2.1 Retail
The retail industry is comprised of retail companies that sell consumer, non-food and pharmaceutical products.
Currently, these companies are in a constant struggle to maintain and increase their profits and to offer the products
and services that the customer (Martinez and Diaz-García, 2021). Many of these companies seek to stand out by means
of a differential factor based on a price approach or by providing personalized service to the consumer and to achieve
this they require an analysis of their buyers and the interaction they have with their products, in order to propose
strategies that allow them to make effective and efficient commercial decisions.
2.2 Market basket analysis
According to Kumar, Kashyap and Gayathri (2021), Market Basket Analysis or (MBA), also known as association
rule learning or affinity analysis, is a data mining technique that can be used in many domains.
Particularly, in the study sector, MBA allows the retailer to understand the buyer's buying behavior, which can help
them in proper decision making and have a new way to understand the variation in data (Kumar et al. 2021). Currently,
this technology is in its infancy within the study sector, which denotes a whole new range of possibilities.
2.3 Rules of association
According to Christian et al. (2021), in a market basket analysis, items A and B have a frequency together, to find the
most representative relationship, the criteria of support, confidence and elevation should be used to measure the
association rules.
© IEOM Society International 2026
Proceedings of the First Australian International Conference on Industrial Engineering and Operations
Management, Sydney, Australia, December 20-21, 2022
Support is used to express the proportion of each identified association (1), confidence refers to the probability that
customers who receive product group "A" will also receive product group "B", measuring the strength of the
relationship between the items (2) and elevation indicates the relationship between the support and the expected one
if A and B are independent (3). In equation (1), N represents the amount of data.
frequency(A, B)
Support (A ⟹ B) = (𝟏𝟏)
N
frequency(A, B)
Confidence (A ⟹ B) = (𝟐𝟐)
frecuency(A)
confidence(A, B)
Lift (A ⟹ B) = (𝟑𝟑)
support (B)
2.4 Apriori Algorithm
One of the data regression approaches is based on employing Machine Learning methodology, which consists in the
use of a priori algorithms based on finding time series regressions within the databases of different fields, since it can
be applied to many sectors, such as: industrial, nuclear, medicine and many others (Pavlyshenko, 2019).
According to Maske and Joglekar (2018), the apriori algorithm allows generating sets conformed by recurrent
elements, thus expanding from set to set until it no longer finds, more valid and successful extensions.
3. Methods
The following research will be quantitative explanatory applied because it will seek to analyze the data obtained from
retail companies and provide a description of their status before, during and after the use of apriori algorithms to
optimize sales and identify products with purchase potential. It is also applied, since it seeks to identify products with
potential to increase the portfolio of groups of articles or products, considering the number of associations obtained.
Python software will be used as tools for the simulation and the application of Machine Learning techniques. The
variables used in this research are the items sold, the transactions carried out and the groups of items.
With respect to the data, it should be noted that to carry out a Market Basket Analysis, the transactions made by
customers should be collected, which allows identifying which products are the most demanded when analyzing
purchasing patterns through the identification and formation of essential associations between products (Christian et
al, 2021).
The hypotheses proposed considering the information collected are the following:
H1: The Machine Learning methodology will allow the identification of products through the association rule, which
will lead to the detection of new related products.
H2: The analysis of historical consumer data will facilitate the proposal of successful commercial strategies.
H3: The application of Machine Learning methodology will optimize the analysis of products and trends in the retail
sector.
4. Data Collection
Considering the above, the data used for the research are transactional data from a supermarket chain of the retail
sector recognized in Latin America that has chains of hypermarkets and supermarkets; specifically, the article focused
on the purchase records of specific stores located in Lima, Peru. For this study the original name of the company will
be omitted for confidentiality reasons, therefore the name Company Alpha will be used. The transactions extracted as
data are from a store representative of the location where the company has operations. It should be noted that we
worked with data from the year 2021.
The data first went through the collection stage, then it went through a series of validation processes and data cleaning
to be used within the algorithms proposed in this report.
© IEOM Society International 2027
Proceedings of the First Australian International Conference on Industrial Engineering and Operations
Management, Sydney, Australia, December 20-21, 2022
Entering in greater detail to the information used for running the algorithms, this went through a statistical sieve, to
eliminate values that were not related to the products sold within the company, since within the transactions there was
evidence of services that would not be useful for the study as ticket rounding, discounts, etc.
Also, to simplify the variables of the report, the different presentations and weights of the products were
conglomerated in groups of articles that would allow better results in the association rules. Consider that the process
through which the data passes is represented in Figure 1.
Figure 1. Processing data
5. Results and Discussion
5.1 Numerical Results
According to the information collected and in the first graphs of the algorithm, it is possible to observe the products
with the greatest support in sales within the period of time that comprise the historical data available, they are grouped
in baskets of up to three products. In particular, in Table 1 it is possible to validate that the products with the greatest
purchase support are basic necessities such as water and vegetables. Particularly, this table is confirmed by baskets of
combinations of a single product.
Taking into account the results in Table 2, it can be seen that the combinations of items with the highest purchase
support are those that take the individual products shown in Table 1.In greater detail, Table 2 takes into consideration
baskets of combinations of up to two products.
Table 1. Sales support per item
Support Itemset
0.3359 {'PLASTIC BAGS'}
0.1220 {'PACKAGED BROWN EGG'}
0.1135 {'PREPARED VEGETABLES'}
0.0989 {'GRASSES AND LEAVES'}
0.0986 {'STILL WATER'}
Table 2. Sales support by item groups
Support Itemset
0.0446 {'VEGETABLE SOUP/ SALAD', 'HERBS AND LEAVES'}
0.0446 {'PACKAGED BROWN EGG', 'PLASTIC BAGS'}
0.0441 {'VEGETABLE SOUP/ SALAD', 'PREPARED VEGETABLES'}
0.0419 {'HERBS AND LEAVES', 'PREPARED VEGETABLES'}
0.0404 {'PLASTIC BAGS', 'PREPARED VEGETABLES'}
© IEOM Society International 2028
Proceedings of the First Australian International Conference on Industrial Engineering and Operations
Management, Sydney, Australia, December 20-21, 2022
5.2 Graphical Results
Continuing with the graphical part of the results achieved, you can see in Figure 2 all the associations made by the
algorithm throughout the entire period of time covered by this investigation and the supports and trusts of each of
them are taken as bases. these.
Figure 2. Total association rules
In order to identify the associations with the highest probabilities and repetitions within the entire algorithm, filters
were applied, which will be mentioned in the following section. The graphical results achieved can be seen in Figure
3. The Figure shows graphically those associations that are most likely to arise or with the highest rates of chances to
occur.
Figure 3. Filtered Association Rules.
Figure 4 is a network graphic, created in the NetworkX library program, used to identify the top rules taking the lift
of each of these as the weighting variable. This graph creates a set of antecedents and consequences after sorting the
total number of rules. Considering that a priori models are unsupervised techniques, it is very important to simplify
© IEOM Society International 2029
Proceedings of the First Australian International Conference on Industrial Engineering and Operations
Management, Sydney, Australia, December 20-21, 2022
the analyst's interpretation experience to make better decisions. In the graph below lift value appear between items
connections.
Figure 4. Network graph with highest lift values
5.3 Proposed Improvements
To carry out the selection of the associations with the best results, filters were applied, where the support of the
antecedents and consequents are greater than 0.02 and 0.01, respectively, as well as the confidence and elevations of
the associations must be greater than 0.45 and 1, respectively. These modifications are represented in Figure 1 and the
best association rules are represented in Table 3.
Table 3. Best Association Rules
Ant Cons
Antecedents Consequents Support Confidence Lift
support support
{'GRASSES AND LEAVES', {'VEGETABLE
0.022 0.097 0.016 0.738 7.574
'CARROTS/GRAINS'} SOUP/ SALAD'}
{'TUBERCLES AND
{'VEGETABLE
ROOTS', 0.024 0.097 0.017 0.713 7.318
SOUP/ SALAD'}
'CARROTS/GRAINS'}
{'CARROTS/GRAINS',
{'VEGETABLE
'TOMATOES AND 0.024 0.097 0.017 0.712 7.303
SOUP/ SALAD'}
PEPPERS'}
{'VEGETABLE SOUP/
{'TUBERCLES AND
SALAD', 'ONIONS AND 0.025 0.082 0.017 0.669 5.107
ROOTS'}
PEPPER'}
The table can be interpreted as the association rules that are most likely to be raised. This can be seen due to their high
confidence indices between combinations of groups of items. Based on these results, future decisions could be made
that affect strategies against the competition.
© IEOM Society International 2030
Proceedings of the First Australian International Conference on Industrial Engineering and Operations
Management, Sydney, Australia, December 20-21, 2022
5.4 Source ode
Next, the source code used in the Python software will be detailed in order to analyze the collected data. As can be
seen, the code was used to analyze only baskets of combinations of 3 groups of items, likewise, a minimum support
of 0.006 was taken.
frequent_itemsets = apriori(df,
min_support = 0.006,
max_len = 3,
use_colnames = True)
frequent_itemsets.head(500)
Due to the amount of data and to present a table detailing the associations with a support greater than 0.01, the
following code was proposed.
rules = association_rules(frequent_itemsets,
metric = 'support',
min_threshold=0.01)
Finally, with the data already displayed, it was possible to identify the best associations through the application of
filters in the support of the antecedents and consequent, as well as in the confidence of each association identified by
the algorithm.
filtered_rules = rules[(rules['antecedent support'] > 0.02)and
(rules['consequent support'] >0.01) and
(rules['confidence'] > 0.01) and
(rules['lift'] > 1.0)]
5.5 Discussion
For future research, it is proposed to test the effectiveness of the portfolio extension and commercial strategies taken
based on the information obtained from the MBA such as customized promotions, cross-selling and space
management.
5.1 Portfolio extension
As can be seen in Table 3, the products with the strongest associations are those of first necessity, emphasizing
vegetables. However, as a retail company, CompanyAlpha sells its vegetables as a single unit and generally does not
have brands on them, so it would not be of much benefit for the company to increase the brands of this group.
As an alternative measure, the data was filtered again, emphasizing the products that are characterized by having
commercial brands, whether they are own, national or imported brands, this new data can be seen in Table 4.
Table 4. Best groups of products with brands per support
Support Itemset
0.3359 {'PLASTIC BAGS'}
0.1220 {'PACKAGED BROWN EGG'}
0.0986 {'STILL WATER'}
0.0629 {'CANNED FISH.TUNA'}
0.0624 {'PACKAGED HAM'}
© IEOM Society International 2031
Proceedings of the First Australian International Conference on Industrial Engineering and Operations
Management, Sydney, Australia, December 20-21, 2022
0.0538 {'FAMILY YOGURT FUNC'}
0.0533 {'FAMILY YOGURT REG'}
0.0529 {'REGULAR BUTTER'}
5.2 Commercial strategies
Customized promotions
According to Moldenhauer and Zwirnmann (2019), analyzing the shopping basket of each buyer allows segmenting
customers according to their demographic traits, life stage or place of residence. Additionally, it is possible to identify
and design new promotional discounts or pricing.
Accordingly, knowing the associations of groups of items allows companies to customize their promotional and
advertising actions depending on the target groups.
Based on the results obtained from the apriori algorithm and using the lift or better known as elevation as the main
indicator. By obtaining a lift greater than 1, it can be affirmed that there is a strong association between the antecedent
and the consequent. A strategy that should be applied is to apply discounts in the groups of items that present a higher
lift, as an example, according to Table 5, would 'EXTRA-GDO 1 RICE, 'CONSERVA PESCAD.TUNA', 'WHITE
BREAD WITH CRUST', among others.
Table 5. Association Rules.
Antecedents Consequents Ant Cons Support Confidence Lift
support support
{'RICE EXTRA-GDE 1'} {'BLOND SUGAR'} 0.043 0.031 0.010 0.240 7.636
{'CANNED FISH.TUNA'} {'LONG NOODLES'} 0.063 0.036 0.011 0.179 4.909
{'RICE EXTRA-GDE 1'} {'ONION OILS AND CHILI PEPPERS'} 0.043 0.050 0.010 0.237 4.710
{'CANNED FISH.TUNA'} {'RICE EXTRA-GDE 1'} 0.063 0.043 0.012 0.196 4.586
{'WHITE BREAD W/CRUSTS'} {'PACKED HAMS'} 0.051 0.062 0.013 0.251 4.017
{'CANNED FISH.TUNA'} {'CRACKERS'} 0.063 0.047 0.011 0.174 3.711
{'VEGETALES OILS'} {'PACKED BROWN EGG'} 0.028 0.122 0.012 0.416 3.410
Cross selling
According to Istrat, V., and Lalić, N. (2017), the identification of associations allows evaluating the implementation
of cross-selling or cross-selling.
Applying a cross-selling strategy allows companies to reduce high inventories, these can be between the same brand
or different brands.
Based on the results obtained in the table 5, companies could implement cross-selling between 'RICE EXTRA-
GDO 1' and 'BLONDE SUGAR', 'CANNED FISH.TUNA' and 'LONG NOODLES', 'CANNED FISH.TUNA' and
'SALT BISCUITS', among others.
Space management
According to Musalem et al (2018), the analysis of shopping baskets allows identifying guidelines for the design of
the distribution of store spaces and when the objective is to reduce the turnover time of products in the store or
minimize the time that the customer takes to make purchases, products should be located based on the probability of
joint purchase of customers.
Retail companies must consider the strongest associations of the products presented by their customers so that the
latter have a pleasant experience and drive the purchase of the products together.
© IEOM Society International 2032
Proceedings of the First Australian International Conference on Industrial Engineering and Operations
Management, Sydney, Australia, December 20-21, 2022
Using the data in the Table 5, the following are proposed to be placed on nearby shelves, considering the lift they
present:
• 'WHITE BREAD WITH RINKS' and 'PACKAGED HAM'
• 'VEGETABLE OILS' and 'PACKAGED BROWN EGG'
• 'BLOND SUGAR' and 'PACKED BROWN EGG'
• 'CANNED FISH.TUNA' and 'LONG NOODLES'
6. Conclusion
O1: Propose the increase of the portfolio of new national and imported products in the study sector.
It can be concluded that, although the products with turnover rates and sales flow are the products of first need,
however, many of these are not characterized by being marketed under a specific brand, but rather under individual
units.
O2: Identify strategies that increase sales and generate healthy margins in companies in the retail sector.
It is highly important to recognize the usefulness of the algorithm to automate the process of identifying potential
sales, since this allows not only to speed up this process, but also to make commercial decisions for the elaboration of
new effective strategies.
O3: Interpret and compare the main statistical indicators of the A priori Algorithms
It can be concluded that, although it is true that support represents a key indicator in the structuring of all association
rules, this is not a factor with which just having it is enough to make trading decisions.
References
Chandwani, M. Market basket analysis using association rule. International Journal of Advance Research, Ideas and
Innovations in Technology, 3-4, 2018.
Christian, M. A., Nathanael, N., Mauliani, A., Indrawati, A., Manik, L. P., and Akbar, Z. Re-al Market Basket Analysis
using Apriori and Frequent Pattern Tree Algorithm. The 2021 International Conference on Computer, Control,
Informatics and Its Applications,161- 165, 2021.
Istrat, V., and Lalić, N. Creating a Decision-Making Model Using Association Rules. Applied Artificial Intelligence,
538–553, 2017.
Kim, R. Y. The impact of COVID-19 on consumers: Preparing for digital sales. IEEE Engineering Management
Review, 2-11, 2021.
Kumar, D., Kashyap, R., and Gayathri, N. Market basket analysis: Identify the changing trends of market data using
association rule mining. Annals of the Romanian Society for Cell Biology, 51-59, 2021.
Marcos Martinez, B., and María García-Díaz, D. Market basket analysis with association rules in the retail sector
using Orange. Case Study: Appliances Sales Company. CLEI Electronic Journal, COLOCAR PAGINAS, 2021
Maske, A., and Joglekar, B. Survey on Frequent Item-Set Mining Approaches in Market Basket Analysis. 2018 Fourth
International Conference on Computing Communication Control and Automation (ICCUBEA), 1-7, 2018.
Moldenhauer, C., and Zwirnmann, H. Basket Analysis in Practice: Mathematical Models and Applications in Offline
Retail. In Performance Management in Retail and the Consumer Goods Industry, 369-384, 2019.
Musalem, A., Aburto, L., and Bosch, M. Market basket analysis insights to support category management. European
Journal of Marketing, 4-20, 2018.
Biographies
Gonzalo Manuel Infante Caldas is a candidate to receive the title of industrial engineer from the Faculty of
Engineering and Architecture of the University of Lima, Lima, Peru.
Xiomara Dayana Molina Rubio is a candidate to receive the title of industrial engineer from the Faculty of
Engineering and Architecture of the University of Lima, Lima, Peru.
Garcia-Lopez Yvan Jesus is PhD (c) in Engineering and Environmental Science, UNALM, “Master of Business
Administration” from Maastricht School of Management, Holland, and master’s in strategic business administration
from Pontificia Universidad Católica del Perú. "Master of Science" in Computer Science, Aerospace Technical Center
© IEOM Society International 2033
Proceedings of the First Australian International Conference on Industrial Engineering and Operations
Management, Sydney, Australia, December 20-21, 2022
- Technological Institute of Aeronautic, Brazil. Stage in Optimization of Processes and Technologies, University of
Missouri-Rolla, USA, and Chemical Engineer from the National University of Callao. Specialization Study in Digital
Transformation, by Massachusetts Institute of Technology, Business Analytics, Wharton School of Management, Data
Science by University of California, Berkeley, Big Data and Data Scientist by MITPro, USA Postgraduate Professor:
Specialized Master from IT, MBA Centrum Católica, MBA from Calgary, Canada, and Centrum Católica. Principal
Consultant DSB Mobile, Executive Director of Optimiza BG, advisor to the Office of Electronic Government and
Information Technology (ONGEI) - PCM, Managing Director of Tekconsulting LATAM, Executive Director of
Optimiza Business Group, Ex- Vice Dean of Information Engineering of the Universidad del Pacifico, Former
Information Technology Manager of “MINERA CHINALCO PERU” Subsidiary of the Transnational Aluminum
Corporation of China, Beijing, China. Former Manager of Systems and Communications of Maple Energy PLC,
Director of Information Technology of Doe Run Peru SRL, Project Manager in implementation of ERP SAP,
EBusiness Suite - Oracle Financial and PeopleSoft. Process Analyst in transnational companies Fluor Daniel
Corporation-USA, PETROBRAS-Brasil, Petróleos del Perú. He has more than 25 years of extensive experience in the
management of investment projects, execution, and commissioning in Peru, Colombia, USA, Brazil, China.
José Antonio Taquía is a Doctoral Researcher from Universidad Nacional Mayor de San Marcos and holds a Master
of Science degree in Industrial Engineering from University of Lima. He is a member of the School of Engineering
and Architecture teaching courses on quantitative methods, predictive analytics, and research methodology. He has a
vast experience on applied technology related to machine learning and industry 4.0 disrupting applications. In the
private sector he was part of several implementations of technical projects including roles as an expert user and in the
leading deployment side. He worked as a senior corporate demand planner with emphasis on the statistical field for a
multinational Peruvian company in the beauty and personal care industry with operations in Europe and Latin
America. Mr. Taquía has a strong background in supply chain analytics and operations modeling applied at different
sectors of the industry. He is also a member of the Scientific Research Institute at the Universidad de Lima being part
of the exponential technology and circular economy groups. His main research interests are on statistical learning,
predictive analytics, and industry 4.0.
© IEOM Society International 2034
Proceedings of the First Australian International Conference on Industrial Engineering and Operations
Management, Sydney, Australia, December 20-21, 2022
© IEOM Society International 2035
