Predictive model based on Machine Learning for the prevention of overstock in a footwear company

Authors

  • Aguirre Méndez, Karina Mercedes
  • Moreno Torres, Alfonso Lorenzo
  • Ovalle, Christian

DOI:

https://doi.org/10.18687/LACCEI2023.1.1.216

Keywords:

Predictive model, Machine Learning, algorithm, overstock, Crisp-DM.

Abstract

Sales forecasting is an essential process for companies as it allows them to plan and make appropriate decisions about their workforce, cash flow and resources. The goal of the sales forecast is to predict future sales based on sales information from the prior period. For the minority sector, it is very important as accurate forecasts can help companies maximize their investments, reduce inventory costs, increase sales and profits, and reduce risks. The most recent and effective method for forecasting future data is Machine Learning. Likewise, in the present work the logistic regression algorithm and decision tree have been applied to determine the best-selling products and categories. The logistic regression algorithm was 97% accurate, with a confusion matrix of 98.1% and 94.4% of true positives and true negatives, respectively. The accuracy metric was 97% and the completeness metric was 96%. The decision tree algorithm was 85% accurate, with a confusion matrix of 86.6% and 83% true positives and true negatives, respectively. The precision metric was 87% and the completeness metric was 84%. It was possible to determine that the ballerina category is the most sold with 84.3%, and that the spring and summer seasons are the most sold.

Downloads

Published

2024-04-16

Issue

Section

Articles