Sustainable Design of a Closed-Loop Automotive Spare Parts Supply Chain Using Machine Learning–Based Product Return Rate Forecasting

Abstract

Faced with mounting ecological issues alongside tighter financial constraints, businesses are turning toward circular supply models that handle goods moving both ways - outward to customers, backward from returns. One of the key challenges in designing such supply chains is accurately forecasting product return rates, as it plays a critical role in location, operational, and capacity planning decisions. In this study, a two-stage framework is proposed based on real-world data obtained from a company operating in the automotive spare parts industry, with a specific focus on brake pads as the target product. In the first stage, product return rates are predicted using three machine learning algorithms—Ridge Regression, Random Forest, and Support Vector Regression—based on six key customer- and order-related features. The model evaluation results indicate that the Random Forest algorithm outperforms the other two models, achieving lower Mean Absolute Error (MAE) and a higher coefficient of determination (R²), thereby providing more accurate predictions. In the second stage, the output of the forecasting model is incorporated as input into a weighted multi-objective mathematical programming model that simultaneously considers economic, environmental, and social objectives for the design of a Closed-Loop Supply Chain (CLSC) network, encompassing eight facility types. Solving the model with real-world data demonstrates that more accurate return-rate forecasts lead to a 16.5% reduction in operational costs and a 33.7% decrease in environmental impacts compared to models with lower predictive accuracy. These results highlight that integrating machine learning techniques with mathematical optimization can significantly enhance the effectiveness of strategic decision-making in CLSC design.