Open AccessEvaluation of Machine Learning Models for Estimating Sales in Physical Retail
Author(s) -
Geovanne Oliveira Alves,
Jorge Lyra,
Alexandre Magno Andrade Maciel
Publication year - 2021
Language(s) - English
Resource type - Conference proceedings
DOI - 10.5753/kdmile.2021.17459
Subject(s) - computer science , confusion matrix , decision tree , sales management , confusion , machine learning , metric (unit) , classifier (uml) , artificial intelligence , marketing , business , psychology , psychoanalysis
The amount of sales in a store is a strong indicator that contributes to managers' decision making. In physical retail, unlike e-commerce, it is more difficult to collect sales and customer behavior metrics because it depends on great sensing and integration between systems. In a shopping mall scenario, we use real WiFi data, People Flow and Sales create a dataset. In this article we propose an evaluation of machine learning models with the objective of estimating the next hour sales in Low, Medium and High, thus providing a tool to assist in decision making. We use the PyCaret library to perform the training of the 13 compared algorithms. The F1-score metric was used to evaluate the models. The Gradient Booster Classifier was the model that got the best result with a score of 84.75%. Among the estimated classes, the High class showed the greatest error in the confusion matrix, reaching 60%, possibly a reflection of the low amount of records in the high class.