A Comparison of the Application of the Exponential Smoothing, Arima, and Random Forest Regression Methods for Forecasting Main Distribution Unit (MDU) Material Requirements at PT XYZ

Abstract

This study aimed to compare the performance of forecasting methods for MDU requirements at PT XYZ using weekly aggregated data. The materials analyzed included kWh meters and Miniature Circuit Breakers (MCBs). Three forecasting methods were compared: Exponential Smoothing as a deterministic approach, ARIMA as a stochastic time-series approach, and Random Forest Regression as a contemporary machine learning-based approach. ARIMA was selected because preliminary identification indicated that the weekly demand data did not exhibit strong seasonal patterns; therefore, a non-seasonal model was considered more appropriate. This research employed a quantitative approach using historical material usage data. The forecasting performance of each model was evaluated using forecasting error metrics, including Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Root Mean Square Error (RMSE). Furthermore, statistical comparisons of forecasting errors were conducted to determine whether performance differences among the methods were statistically significant. The best-performing forecasting model was then used as the basis for developing preliminary recommendations for inventory policy. This study is expected to contribute scientifically to the selection of appropriate forecasting methods based on the characteristics of MDU demand data. In addition, it provides practical contributions for PT XYZ by supporting more objective, measurable, and data-driven procurement planning and inventory management.