Demand Forecasting in Micro-Fulfillment Centers Using Association Rule-Based Machine Learning

Abstract

The rise of Micro-Fulfillment Centers (MFCs) has intensified the need for accurate and interpretable demand forecasting. To address the irregular and volatile nature of MFC demand, we propose Association Rule-based Machine Learning (ARML), a novel approach that integrates association rule mining with machine learning for enhanced feature selection and predictive accuracy. This study goes beyond simple accuracy comparisons by conducting statistical validation and systematically evaluating ARML across multiple experimental setups that account for varying levels of demand volatility and operational complexity. Using parcel delivery data from South Korea, we identify key conditions under which ARML excels, demonstrating its robustness and practical implications for real-world MFC operations. Furthermore, by leveraging association rules as an explainability mechanism, ARML contributes to the field of eXplainable AI (XAI). The results confirm that ARML consistently outperforms benchmark models, providing a scalable, interpretable, and high-performing solution for dynamic e-commerce logistics.

Keywords: Forecasting, Machine learning, Association Rules, Micro-Fulfillment Center, Explainable AI, e-commerce logistics