A Machine Learning-based Framework to Configure a Sustainable Reverse Logistics in the Steel Industry
Ali Dosti
1
(
Department of Industrial Management, Qazvin Branch, Islamic Azad University, Qazvin, Iran
)
Arash Hajikarimi
2
(
Department of Management, Faculty of Humanities٫Raja University, Qazvin, Iran
)
Sadegh Abedi
3
(
Department of Industrial Management, Qazvin Branch, Islamic Azad University, Qazvin, Iran
)
Keywords: Supply chain management, Reverse logistics, Sustainability, Machine learning-based model, Steel industry,
Abstract :
The steel industry plays a pivotal role in modern industrial development but also poses significant sustainability challenges due to its high levels of waste and environmental impact. This study presents a comprehensive data-driven framework for designing a sustainable reverse logistics network in the steel industry, incorporating the three pillars of sustainability—economic, environmental, and social. To achieve this, a multi-objective mathematical programming model is proposed with the aim of maximizing total profit and social benefits while minimizing environmental damage. To address uncertainty, a hybrid approach is introduced by integrating Robust Possibilistic Programming (RPP) with the Gradient Boosting algorithm, which allows for accurate prediction of waste generation using key process features such as furnace efficiency, process temperature, air conditions, and operator experience. A dataset of 2,000 production records was used to train and evaluate the machine learning model, achieving high prediction accuracy (R² = 0.96, RMSE = 0.31%). The optimization model is solved using the General Revised Multi-Choice Goal Programming (GRMCGP) method, which enables the incorporation of multiple aspiration levels and balances the trade-offs among conflicting objectives. A real-world case study from a steel plant in southern Iran demonstrates the practical applicability of the model. The results show that increasing recycling capacity leads to a rise in overall profits while simultaneously reducing environmental and social impacts. Additionally, sensitivity analyses provide deeper insights into how key parameters—such as initial waste volume and facility capacity—affect the model’s performance. Managerial implications are discussed, emphasizing how the proposed framework can support strategic decision-making by providing actionable insights under uncertainty. This study contributes to both academic literature and industry practice by presenting a novel, scalable, and flexible machine learning-driven optimization framework for configuring sustainable reverse logistics in steel manufacturing.