Page 1 of 33
Archives of Business Research – Vol. 12, No. 6
Publication Date: June 25, 2024
DOI:10.14738/abr.126.17093.
Alves, G. O., Maciel, A. M. A., Fonseca, J. C. B., Medeiros, E. C., Moser, P. C., de Andrade, R. C. D., de Carvalho, F. F., Junior, F. P. de
S. L., & Domingues, M. A. de O. (2024). Unlocking Retail Success: Empowering Decision-Making with Advanced Sales Forecast
Models. Archives of Business Research, 12(6). 12-44.
Services for Science and Education – United Kingdom
Unlocking Retail Success: Empowering Decision-Making with
Advanced Sales Forecast Models
Geovanne Oliveira Alves
ORCID: 0000-0002-2084-5516
University of Pernambuco, Caruaru, PE, Brazil
Alexandre Magno Andrade Maciel
ORCID: 0000-0003-4348-9291
University of Pernambuco, Recife, PE, Brazil
Jorge Cavalcanti Barbosa Fonseca
University of Pernambuco, Caruaru, PE, Brazil
Erika Carlos Medeiros
ORCID: 0000-0003-2506-7116
University of Pernambuco, Caruaru, PE, Brazil
Patricia Cristina Moser
University of Pernambuco, Caruaru, PE, Brazil
Rômulo César Dias de Andrade
University of Pernambuco, Caruaru, PE, Brazil
Fernando Ferreira de Carvalho
University of Pernambuco, Caruaru, PE, Brazil and Federal Institute
of Science and Technology of Pernambuco, Recife, PE, Brazil
Fernando Pontual de Souza Leão Junior
University of Pernambuco, Caruaru, PE, Brazil
Marco Antônio de Oliveira Domingues
ORCID: 0000-0002-7579-3485
Federal Institute of Science and Technology of Pernambuco, Recife, PE, Brazil
ABSTRACT
The gross revenue indicator contributes to the understanding of the company’s
situation, and generating sales revenue forecasts is a strategy that helps the
manager in directing the business. This work aims to develop a set of Machine
Learning (ML) models to forecast sales in physical retail. Methodology – To carry
out this work, a methodology was proposed to create, compare and evaluate ML
models. Findings – When analyzing the forecast scenarios, it was observed that
Hourly forecasts performed better than Day forecasts. We highlight the LIGHTGBM
model, which presented the best scores in the F1-score metric with 82.95%, 79.26%
Page 2 of 33
13
Alves, G. O., Maciel, A. M. A., Fonseca, J. C. B., Medeiros, E. C., Moser, P. C., de Andrade, R. C. D., de Carvalho, F. F., Junior, F. P. de S. L., & Domingues,
M. A. de O. (2024). Unlocking Retail Success: Empowering Decision-Making with Advanced Sales Forecast Models. Archives of Business Research,
12(6). 12-44.
URL: http://doi.org/10.14738/abr.126.17093
and 76.53% scenario representing one hour, two hours and three hours ahead,
respectively. Value – It is expected that the forecast models will help managers to
find insights to support the operational decisions of physical retail contributing to
carry out actions to optimize companies’ processes.
Keywords: Machine Learning, Physical Retail, Sales Forecasting, Shopping Mall.
INTRODUCTION
The retail sector has witnessed significant growth in recent years, even amidst the challenges
posed by the Covid-19 pandemic. According to the Sociedade Brasileira de Varejo e Consumo
(SBVC), the restricted retail sector, encompassing consumer goods retail excluding automobiles
and construction materials, recorded a growth rate of 1.2% in 2020. while GDP experienced a
4.1% decline. With a total turnover of R$1.75 trillion, this sector constitutes a substantial 23.6%
of the Brazilian GDP. This resilience and transformative potential of the retail industry
underscore the importance of understanding and predicting key performance indicators to
steer business strategies effectively [1]. One of the critical performance indicators in the retail
domain is gross revenue, representing the total income generated through sales. Effective
forecasting of sales revenue can provide valuable insights to managers, aiding in decision- making, resource allocation, and optimizing operational processes. Sales forecasting has
become even more crucial in the dynamic retail landscape, where the interplay of physical
stores and online sales channels necessitates accurate predictions to seize market
opportunities, enhance customer experience, and outperform competitors [2].
The emergence of digital technologies and the growth of e-commerce have further shaped the
retail landscape. Companies are increasingly investing in online sales methods to overcome
geographical barriers and tap into diverse consumer segments. E-commerce enables the
collection of valuable data on customer behavior, including product viewing time, clicks, cart
contents, and sales patterns. Structuring and leveraging this data through artificial intelligence
systems facilitate the generation of predictions and recommendations, boosting sales, and
fostering customer loyalty [3]. Despite the growing importance of e-commerce, physical stores
remain a crucial aspect of the retail ecosystem. Shopping malls, in particular, house various
stores catering to diverse audiences, making data collection and analysis more complex.
Extracting relevant sales data and customer behavior from physical retail settings becomes
challenging due to the lack of streamlined process monitoring. To address this, our work aims
to develop a set of Machine Learning models for sales forecasting in physical retail, particularly
in shopping malls.
The objective of this research is to create, compare, and evaluate Machine Learning models that
can forecast sales revenue in physical retail settings. We will explore different scenarios and
examine the performance of these models in hourly and daily forecasts. By accomplishing this
objective, we seek to empower retail managers with valuable insights to support strategic and
operational decisions, optimize workforce allocation, improve inventory management, and
develop targeted advertising campaigns, among other actionable strategies to thrive in the
competitive retail market. Through this research, we aspire to bridge the gap between digital
and physical retail by providing an effective data-driven approach for sales forecasting in
traditional brick-and-mortar stores.
Page 3 of 33
14
Archives of Business Research (ABR) Vol. 12, Issue 6, June-2024
Services for Science and Education – United Kingdom
General Objective
The objective of this research is to develop a set of Machine Learning models to forecast sales
revenue in physical retail.
Specific Objectives
1. Conduct a survey of the state of the art in the area of data mining;
2. Create a dataset with sales information and target audience behavior in physical stores;
3. Compare and analyze the performance of Machine Learning classification models to
predict sales revenue.
RELATED WORKS
Triayudi et al. [4] utilized time series methods to predict future transaction values based on
historicalsales data. However, they only used approximately 5000 records per year from the
sales database for training and validation, potentially underutilizing the available data. The
results were satisfactory, achieving a high precision level of 99.68%, providing valuable
estimates and visual patterns forbusiness process management. In the realm of e-commerce,
Gonçalves [5] conducted a comparison between multiple regression techniques and time series
for cost and revenue forecasting in a virtual store. The analysis focused on key performance
indicators, utilizing data from Facebook Ads and Google Ads, such as clicks, impressions, reach,
conversions, and items added to the cart. The multiple regression technique out- performed
time series, achieving 98% accuracy in cost forecast and 68% in revenue forecast based on the
adjusted R-squared metric.
Huber et al. [6] emphasized the importance of forecasting retail demand in operational
decisionmaking, particularly on special days like holidays, which exhibit distinct behavioral
patterns. Their research targeted a chain of bakeries, predicting demand for common bakery
products such as rolls, breads, pastries, cakes, and snacks. They evaluated different methods,
including artificial neural networks and decision trees with gradient boosting, finding that
classification-based approaches outperformed regression-based approaches.
Fildes et al. [7] conducted a comprehensive literature review on forecasting retail demand,
categorizing the challenges faced by retailers at strategic, tactical, and operational levels.
Strategic decisions demand long-term forecasts that consider environmental and technological
changes, while tactical decisions focus on medium-term profitability and operational decisions
on day-to-day activities, influencing the success of higher-level decisions.
Pundir et al. [8] developed an ML application to predict the future revenue generation of a retail
chain, incorporating not only previous sales figures but also non-intuitive external factors such
as economic performance, consumer price index, unemployment rate, and ambient
temperature. They used a database from 45 Walmart stores on Kaggle and concluded that
multivariate time series ML models can enhance forecast accuracy due to the significance of
external and internal variables.
The related works presented in this section demonstrate the importance and effectiveness of
using Machine Learning (ML) techniques for sales forecasting in various retail settings. Building
on the insights gained from these related works, the current study aims to develop a set of ML