Previsão de demanda em uma empresa de produção alimentícia: uma análise comparativa entre métodos básicos e machine learning

Abstract

Demand forecasting is critical for food production companies, directly impacting strategic and operational decisions. This study conducts a comparative analysis between traditional methods (simple moving average and weighted moving average) and machine learning techniques (SARIMA and XGBoost) for demand forecasting. The rationale is to enhance forecast accuracy, optimize inventory, and minimize operational costs. Objectives include assessing the accuracy of each method using metrics such as Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE), and identifying the most suitable method for the company's specific context. The methodology involves applying each model to historical sales data using statistical tools, data preprocessing techniques, and machine learning methods. Preliminary results indicate that machine learning models tend to outperform traditional methods, providing more precise forecasts adaptable to seasonal variations and nonlinear trends. It is concluded that implementing XGBoost can offer significant advantages in demand forecasting for food production companies, enabling better resource management and greater responsiveness to market fluctuations.