A Decision-Intelligence Framework for Explainable and Reliable Inventory Planning

Abstract

Modern supply chain systems require more than accurate demand forecasts and they also need decision making tools that can be used directly in the real operations. While many machine learning models provide strong predictive performance, they often lack in transparency and are difficult to use in the practice. In this work, we present a decision intelligence framework that brings together the prediction, the explanation and in the decision-making for inventory planning. Demand is estimated using a gradient boosting model and a SHAP is used to understand the main factors which are influencing each prediction. These insights are then used to group predictions into simple, context-based decision categories, helping move from model output to actual decisions. The framework also includes a reliability component that estimates how confident each prediction is, allowing actions to be prioritized under uncertainty.