Beyond One-Size-Fits-All: A Demand Classification and Decomposition Approach for Intermittent Demand Forecasting

Abstract

This study evaluated the performance of Croston, Syntetos–Boylan Approximation (SBA), and Teunter–Syntetos–Babai (TSB) methods in forecasting intermittent demand using a demand classification and decomposition approach in the document solutions industry. The analysis was conducted using monthly SKU-level data for 134 items from August 2023 to November 2025. Demand was classified into smooth, intermittent, erratic, and lumpy types based on the Average Demand Interval (ADI) and squared Coefficient of Variation (CV²), and further analyzed by separating demand into total, rental, and sales categories. Forecast accuracy was evaluated using RMSE MAE.The results showed that forecasting performance varied across demand types, demand categories, and evaluation metrics. TSB was more effective in reducing large forecasting errors, while SBA provided more stable performance. These findings indicated that RMSE was more relevant for controlling extreme deviations, such as stockout risks, whereas MAE supported more stable inventory planning. In addition, demand aggregation tended to mask underlying variability, while demand decomposition provided more detailed insights. Overall, the study demonstrated that forecasting should not be applied uniformly across heterogeneous SKUs. A demand-based approach that integrates classification and decomposition enabled more accurate forecasting, reduced inventory risk, and supported more effective operational decision-making.