Logistics optimization using machine learning: Demand forecasting and transport in supply chain

Loading...
Thumbnail Image

Journal Title

Journal ISSN

Volume Title

Publisher

University of Tlemcen

Abstract

Supply chain management faces two persistent challenges: demand uncertainty and transportation inefficiency, which degrade inventory planning, logistics costs, and service levels in large-scale, multi market e-commerce. This thesis proposes a machine learning-based demand forecasting framework using real data from the DataCo Global Supply Chain dataset, covering five markets: Africa, Europe, LATAM, Pacific Asia, and USCA. Dedicated models are trained per market to capture regional demand heterogeneity. Three ensemble algorithms Random Forest, XGBoost, and LightGBM are compared using MAE, MSE, and R². LightGBM achieves the best overall performance, particularly in high-variability markets like Pacific Asia. The forecasts support logistics decisions: market prioritization, vehicle allocation, shipping mode assignment, and bottleneck identification. Three extensions are proposed: metaheuristic vehicle routing optimization, a Reinforcement Learning agent for adaptive shipping mode selection, and ERP integration as a forecasting module. This confirms machine learning's potential for proactive, scalable supply chain management.

Description

Citation