Integrating driver behavior into last-mile delivery routing: Combining machine learning and optimization in a hybrid decision support framework

The overall quality of last-mile delivery in terms of operational costs and customer satisfaction is primarily affected by traditional logistics planning and the consideration and integration of driver knowledge and behavior. However, this integration has yet to be exploited. This phenomenon is mirrored in two largely separated research bodies on logistics planning and driver behavior. Bridging this gap by using and integrating historical data from actually driven tours into last-mile delivery planning is promising for research and practice. Still, it also leads to complex and large-scale routing problems, which require the development of an overall methodology that goes beyond classical optimization approaches as the needed approach requires a multi-stakeholder perspective, calls for a hybrid-analytical approach by incorporating tour prediction and prescription, and requires both data science and optimization methods. Accounting for these challenges, we suggest a hybrid decision support framework for the traveling salesman problem with time windows that combines machine learning techniques and conventional optimization methods and considers the deviation between suggested and predicted tours. We demonstrate the applicability of our framework in a case study that draws on real-world logistics data. Relying on a sensitivity analysis, we investigate and illustrate the trade-off between the level of deviation between predicted and suggested tours and tour costs. Our case study draws general managerial implications and recommendations that guide decision makers in building their decision support systems for last-mile delivery routing by instantiating our generic framework.