OPPORTUNITIES AND LIMITATIONS OF MACHINE LEARNING ALGORITHMS FOR DEMAND FORECASTING IN INVENTORY MANAGEMENT

Abstract

Accurate demand forecasting is essential for inventory control, yet increasingly volatile markets and intermittent product demand limit the reliability of conventional forecasting methods. This conference thesis critically examines statistical, ensemble machine learning, and deep learning approaches, focusing on their predictive capabilities, methodological weaknesses, and implications for inventory decisions. The analysis identifies data quality, information leakage, model overfitting, inappropriate evaluation metrics, and the gap between forecast accuracy and inventory costs as major challenges. A reproducible, time-aware comparison framework is proposed to support future empirical research. The contribution is methodological and analytical; no unperformed experiments or unsupported numerical improvements are reported.