Comparative Analysis Between Classical Methods and Artificial Intelligence Approaches in Demand Forecasting

Dorra Dridi, Younes Boujelbene

Demand forecasting is a critical component of supply chain management, directly impacting inventory control, production planning, and overall operational efficiency. This study aims to compare the performance of classical forecasting methods and an artificial intelligence (AI)-based approach in the context of demand forecasting. The classical methods considered include Simple Moving Average (SMA), Weighted Moving Average (WMA), Simple Exponential Smoothing (SES), and linear regression, while the AI-based approach is represented by the LightGBM algorithm. Using a real dataset of 80 observations, a rigorous experimental framework was implemented to evaluate the accuracy and robustness of each method based on three performance metrics: Mean Squared Error (MSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). The results indicate that forecasting performance varies significantly across methods. Among classical techniques, linear regression achieved the best performance, demonstrating its effectiveness in capturing demand trends. However, the LightGBM model outperformed all classical methods across all evaluation metrics, highlighting its superior ability to model complex and nonlinear demand patterns, particularly in dynamic environments. This study contributes to the literature by offering a unified comparative framework and providing practical managerial insights for selecting appropriate forecasting methods. It also opens perspectives for the development of hybrid approaches combining classical and AI techniques to enhance forecasting accuracy and decision-making.