Comparación de Modelos de Series Temporales para el Pronóstico de la Demanda Semanal de Cocinas en una Empresa Manufacturera de la Ciudad de Guayaquil

Abstract

Production planning in manufacturing requires accurate forecasts. This study compares forecasting models for optimizing kitchen production. The objective of this research is to evaluate four forecasting methodologies to determine the most effective for predicting weekly kitchen production. For this purpose, 48 weeks of production data were analyzed using simple moving average, weighted moving average, simple exponential smoothing, and an artificial neural network. Evaluation metrics included MAPE, MAE, RMSE, bias, and R² to objectively compare the predictive performance of each model. The research results demonstrated that the artificial neural network showed significant superiority with a MAPE of 8.52% and an R² of 0.783, representing a 53.8% improvement over the best traditional model. The conventional methods showed limitations in capturing nonlinear patterns, with MAPE values ranging from 18.03% to 21.43%. The ANN reduced the average absolute error to 244 units compared to 501–571 units for the other models. It can be concluded that neural networks constitute the optimal alternative for production forecasting in the kitchen manufacturing industry.