Integrated machine learning model for managing customer-driven end-to-end supply chain uncertainty

Abstract

Supply chain management is increasingly challenged by global disruptions, globalisation, and demand volatility, complicating end-to-end supply chain management in manufacturing. The shift from Industry 4.0 to Industry 5.0 emphasises sustainability, human-centricity, and resilience, highlighting the urgent need to address customer-driven end-to-end supply chain uncertainty. This uncertainty encompasses managing backorder risks and forecasting recycled end-of-life product quantities to promote circularity and sustainability. Current methods fall short in mitigating this uncertainty due to computational limitations. However, machine learning, a key advancement from Industry 4.0, offers a promising solution. Thus, an integrated machine learning model is proposed to mitigate customer-driven end-to-end supply chain uncertainty. This model combines a Bayesian-optimised light gradient-boosting machine (LGBM) to predict backorder risks and a k-nearest neighbour mega-trend diffusion (KNNMTD)-optimised Stacking ensemble model to forecast recycled end-of-life product quantities. Case studies show the Bayesian-optimised LGBM model outperforms benchmark models in accuracy, recall, and AUC (all > 0.8) with 125 s of operational time, while the KNNMTD-optimised Stacking model achieves a superior R2 of 0.9515 compared to baseline models. This integrated model enhances prediction performance, generalisation, and supply chain resilience. It enables manufacturers and remanufacturers to optimise production and remanufacturing, allocate resources effectively, and promote a sustainable, circular value chain.