Explainable, Low-Resource ML-Based Inventory Forecasting for SMEs: Bridging the Gap for Non-Technical Users

Abstract Effective inventory forecasting is essential for small and medium-sized enterprises (SMEs), where stockouts and excess inventory can significantly affect working capital, operating costs, and customer satisfaction. Machine learning methods can improve forecasting by identifying complex patterns in sales and inventory data; however, their practical adoption is constrained by data scarcity, computational and financial limitations, technical complexity, and a lack of interpretable outputs. This paper reviews traditional statistical, tree-based, neural network, hybrid, and explainable artificial intelligence approaches to inventory forecasting, with particular attention to their suitability for resource-constrained SMEs and non-technical users. The analysis indicates that although advanced models may offer strong predictive performance, they are not always appropriate for SMEs because of their computational requirements and limited interpretability. Tree-based and lightweight models may offer a more practical balance among predictive accuracy, computational efficiency, and explainability. The literature also reveals gaps between forecast accuracy and actual inventory outcomes, alongside limited evaluation involving SME users. Future solutions should therefore combine resource-efficient machine learning with clear, decision-focused explanations and evaluate performance using both forecasting metrics and inventory-related outcomes. Such an approach can support more trustworthy, affordable, and practically useful inventory forecasting for SMEs.