Machine Learning-Driven Recommender System for Green Postal Hub Location: Joint Reduction of Fuel Consumption and Operational Costs

This study proposes a machine learning–driven recommender system for green postal hub location planning, aimed at jointly reducing fuel consumption, operational cost, and carbon-related impacts while maintaining service effectiveness. Unlike conventional facility-location formulations that yield a single static solution, the problem is modeled as a zone-to-hub recommendation task in which candidate hubs are ranked for each demand zone. A domain-informed synthetic dataset is constructed to represent realistic postal logistics conditions, comprising ۵۰۰ demand zones and ۱۰۰ candidate hubs and integrating spatial, operational, economic, and sustainability attributes (e.g., distance, demand intensity, congestion, land-cost index, hub capacity, solar potential, and electric-vehicle charging infrastructure). Multiple learning models and ensemble baselines are trained and compared, and a group-aware validation strategy based on zone identifiers is employed to prevent data leakage and to ensure reliable generalization estimates. Performance is assessed using both classification measures and recommender-oriented ranking metrics, including NDCG@K, MAP@K, MRR, and HitRate@K, enabling rigorous evaluation of top K hub recommendations. The results demonstrate that the proposed ranking-based framework can deliver highly accurate hub recommendations (exceeding ۹۰% predictive performance in the constructed setting) while improving the quality of top K selections, and feature-level analyses further indicate the importance of sustainability indicators in the final recommendations.

کلیدواژه ها:

green logistics ، postal hub location ، recommender systems ، learning-to-rank ، multi-objective optimization ، sustainable transportation