Optimización de la gestión del inventario de producto terminado en una fábrica productora de aderezos y salsas, mediante modelos de aprendizaje automático

Abstract

This project addresses the problem of inventory planning in a food manufacturing company through short-term demand forecasting. The objective was to develop a system capable of estimating product sales for the next fifteen days to optimize inventory levels and improve operational efficiency. The proposal is justified by the need to reduce storage costs, prevent stockouts, and strengthen data-driven decision-making. Historical sales data from 2024 were used, including information on quantities sold, prices, dates, and product characteristics. The data underwent cleaning, integration, and feature-enrichment processes, after which three predictive models were trained: Exponential Moving Average, Multilayer Perceptron, and Random Forest. The experiments followed a supervised learning methodology and used error metrics such as MAE and RMSE. The results showed that the Random Forest model achieved the best performance across all product segments. Therefore, it constitutes the proposed solution for the forecasting system. It is concluded that this approach improves inventory planning and provides a reliable tool for decision support. Keywords: Demand forecasting, Random Forest, inventory, data analysis, predictive models.