Smart Predictions: Machine Learning for Demand Forecasting Review and analysis

¹ Laboratory of industrial techniques, faculty of sciences and technology, Fez,Morocco, yahya.hmamou@usmba.ac.ma
² Laboratory of industrial techniques, faculty of sciences and technology, Fez,Morocco,anaschafi@gmail.com
³ Laboratory of industrial techniques, faculty of sciences and technology, Fez,Morocco, salaheddine.kammourialami@usmba.ac.ma

Abstract

Accurate demand forecasting is critical for improving supply chain operations in Logistics 4.0, although traditional statistical approaches struggle to represent the complex, non-linear trends in current demand data. This study analyzes machine learning (ML) applications in demand forecasting across industrial supply chains, evaluating 21 research from 2020 to 2024. We propose a novel classification framework that categorizes ML techniques by algorithm type (e.g., deep learning, ensemble methods) and data characteristics (e.g., volume, dimensionality),showing themes including the advantage of LSTM networks (Long Short-Term Memory) for high-volume, multivariate data. Our findings demonstrate deep learning methods minimize predicting errors compared to standard approaches, while computing needs may restrict implementation. This categorization gives supply chain practitioners a practical guidance to identify appropriate ML approaches, boosting efficiency and profitability in changing marketplaces.