Human-Centered Artificial Intelligence Framework for Adaptive Inventory Decision Support under Supply Chain Uncertainty

Abstract

Supply chain uncertainty has increased the complexity of inventory decision-making due to demand fluctuations, supplier disruptions, and rapidly changing operational conditions. Although artificial intelligence (AI) has demonstrated strong potential in improving supply chain analytics, existing approaches often investigate prediction, explainability, optimization, and human involvement as separate components. This study proposes a Human-Centered Artificial Intelligence (HCAI) framework for adaptive inventory decision support under supply chain uncertainty by integrating AI-based demand prediction, Explainable Artificial Intelligence (XAI), adaptive inventory optimization, and human decision support within a unified framework. The Design Science Research Methodology (DSRM) was adopted to develop and evaluate the proposed framework using empirical inventory data collected from food distributors and rice warehouses in Pidie Regency and Bireuen Regency, Aceh, Indonesia. Forecasting models, including ARIMA, Exponential Smoothing, Random Forest, LSTM, and XGBoost, were evaluated using RMSE, MAE, and MAPE, with the best-performing model subsequently integrated with SHAP-based explanations and a Mixed-Integer Programming (MIP) optimization model. The evaluation results showed that XGBoost achieved the highest forecasting performance with a MAPE of 7.40%. The proposed framework reduced total inventory cost by 22.7%, improved service level from 88.5% to 96.8%, and decreased stockout rate from 11.5% to 3.2% compared with the conventional inventory policy. Furthermore, user evaluation involving 20 inventory practitioners indicated positive acceptance of the Human Decision Support Layer, with high scores for perceived usefulness, trust, and decision confidence. These findings demonstrate that integrating predictive intelligence, explainable AI, adaptive optimization, and human oversight enable transparent and adaptive inventory decision support under uncertain supply chain conditions. This study contributes to advancing Human-Centered AI applications in supply chain management by providing an integrated framework that bridges AI capability with human operational expertise