Bridging operations research and machine learning for service cost prediction in logistics and service industries

Abstract Optimizing shared resources across multiple clients is a complex challenge in the production, logistics, and service sectors. This study addresses the underexplored area of forecasting service costs for non-cooperative clients, which is essential for sustainable business management. We propose a framework that merges Operations Research (OR) and Machine Learning (ML) to fill this gap. It begins by applying the OR model to historical instances, optimizing resource allocation, and determining equitable service cost allocations for each client. These allocations serve as training targets for ML models, which are trained using a combination of original and augmented client data, aiming to reliably project service costs and support competitive, sustainable pricing strategies. The framework’s efficacy is demonstrated in a reverse logistics case study, benchmarked against two traditional cost estimation methods for new clients. Comparative analysis shows that our framework outperforms these methods in terms of predictive accuracy, highlighting its superior effectiveness. The integration of OR and ML offers a significant decision-support mechanism, improving sustainable business strategies across sectors. Our framework provides a scalable solution for cost forecasting and resource optimization, marking progress toward a circular, sustainable economy by accurately estimating costs and promoting efficient operations.