Integrating ensemble machine learning with TOPSIS and VIKOR for inventory waste index prediction and driver prioritization: evidence from Bangladesh

Inventory waste, resulting from both excess stock and stockouts, remains a major challenge in modern supply chains due to demand uncertainty and operational complexity. This study proposes an integrated framework combining ensemble machine learning and multi-criteria decision-making (MCDM) methods to predict and prioritize the key drivers of inventory waste using the Inventory Waste Index (IWI). Multiple models are evaluated through nested cross-validation, where CatBoost demonstrates the highest predictive performance (R 2 = 0.8116), outperforming traditional linear regression. Feature importance analysis identifies the most influential operational variables, which are subsequently prioritized using entropy weighting with TOPSIS and VIKOR ranking methods. The results show strong consistency between the two MCDM techniques and reveal that stockout frequency and overstock value are the dominant contributors to inventory waste. The proposed framework provides a robust data-driven decision-support approach for identifying high-risk inventory drivers and improving inventory management strategies.