Forecasting Return Quantity, Timing and Condition in Remanufacturing with Machine Learning: A Mixed-Methods Approach

Abstract

Remanufacturing plays a key role in the circular economy by reducing material consumption and extending product life cycles. However, a major challenge in remanufacturing is accurately forecasting the availability of cores, particularly regarding their quantity, timing, and condition. Although machine learning (ML) offers promising approaches for addressing this challenge, there is limited clarity on which influencing factors are most critical and which ML approaches are best suited to remanufacturing-specific forecasting tasks. This study addresses this gap through a mixed-method approach combining expert interviews with two systematic literature reviews. The interviews with professionals from remanufacturing companies identified key influencing factors affecting product returns, which were structured into an adapted Ishikawa diagram. In parallel, the literature reviews analyzed 125 peer-reviewed publications on ML-based forecasting in related domains—specifically, spare parts logistics and manufacturing quality prediction. The review categorized data sources into real-world, simulated, and benchmark datasets and examined commonly applied ML models, including traditional methods and deep learning architectures. The findings highlight transferable methodologies and critical gaps, particularly a lack of remanufacturing-specific datasets and integrated models. This study contributes a structured overview of ML forecasting in remanufacturing and outlines future research directions for enhancing predictive accuracy and practical applicability.