Authors:
Jociano Perin
;
Lucas Dias Hiera Sampaio
;
Marlon Marcon
and
André Roberto Ortoncelli
Affiliation:
Postgraduate Program in Informatics, Federal University of Technology, Paraná (UTFPR), Paraná, Brazil
Keyword(s):
Sales Prediction, Supermarket Sales, Time Series Forecasting, Long Short-Term Memory, Gated Recurrent Unit.
Abstract:
Accurately predicting daily sales of products in supermarkets is crucial for inventory management, demand forecasting, and optimizing supply chain operations. Many studies focus on predicting the total sales of large stores and supermarkets. This study focuses on forecasting daily sales of individual products across various categories. In the experiments, we used Linear Regression and two types of Recurrent Neural Networks: Long Short-Term Memory and Gated Recurrent Unit. One of the contributions of the work is the database used, which is made available for public access and contains daily sales records (between January 2019 and December 2024) of 250 products in a medium-sized supermarket in Brazil. The results show that the predictors’ performance varies significantly from product to product. For one semester, the average of the best 25% resulted in a Root Mean Squared Error (RMSE) of 1.55 and a Mean Absolute Percentage Error (MAPE) of 17.20, and for the average of all products, the
best RMSE was 2.12, and the best MAPE was 43.94. We observed similar performance variations for all analyzed semesters. With the results presented, it is possible to understand the performance of the predictors in ten semesters.
(More)