A machine learning approach for marginal fulfillment cost estimation in last mile delivery

This thesis investigates a machine learning model to tackle the complex problem of accurately estimating the marginal fulfillment cost (MFC) for last-mile delivery for the online retail sector. Fast and accurate MFC estimation is essential for effective real-time management of online retail operations. However, due to the high computational complexity of the underlying vehicle routing problems (VRP), it is not possible to directly calculate the marginal delivery costs for real-time demand management and resource allocation decisions. To address this challenge, we propose a novel machine learning (ML) approach that eliminates the need to solve time-consuming VRP instances to calculate MFC and provides accurate predictions in milliseconds. We demonstrate the effectiveness of our proposed methodology through extensive numerical experiments with real-world delivery data. Our results show that the ML model outperforms the state-of-the-art location-based MFC estimation methods, improving decision-making capabilities for online retailers to reduce their costs, improve customer satisfaction, and reduce the negative externalities of delivery operations.