Comparing ARIMA and Single Exponential Smoothing for Spare-Part Demand Forecasting: A Case Study at PT XYZ

Abstract

Purpose: This study aims to compare the forecasting performance of the Autoregressive Integrated Moving Average (ARIMA) and Single Exponential Smoothing (SES) methods using Battery DYC spare-part demand data from PT XYZ as an industrial case study to identify the most appropriate forecasting approach for inventory planning. Methods: Monthly sales data from January 2020 to May 2026 were analyzed using a quantitative time-series approach. The dataset was divided into training data (65 observations) and testing data (12 observations). The ARIMA model was developed according to the Box-Jenkins procedure, while the SES model used the optimal smoothing parameter estimated by SPSS. The forecast performance was assessed in terms of Root Mean Square Error (RMSE), Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE). Result/Findings: The results show that ARIMA (0,0,1) model has RMSE of 3.8370, MAE of 2.8925 and MAPE of 53.21%. SES (alpha = 0.052) has RMSE of 3.8336, MAE of 3.0000 and MAPE of 55.19%. SES is slightly better than ARIMA in terms of RMSE but ARIMA is better in MAE and MAPE, thus ARIMA is more robust overall. Novelty/Originality/Value: The study provides empirical evidence from an industrial case study that forecasting performance is dependent on demand characteristics, rather than on the universal superiority of one method over another. The findings offer practical guidance for spare-part inventory planning and provide a basis for future comparisons with specialized intermittent demand forecasting methods.