Séminaire en format hybride au local 4488 du GERAD ou Zoom.

We present a contextual framework for learning routing experiences in last-mile delivery. The objective of the framework is to generate routes similar to historic high-quality ones as classified by the operational experts by considering the unstructured features of the last-mile delivery operations. The framework encompasses descriptive, prescriptive and predictive analytics. In the descriptive analytics, we extract rules and preferences of high-quality routes from the data. In the predictive analytics stage, we investigate different derivative-free algorithms for learning the preferences in order to improve the effectiveness of the methods. We develop a label-guided algorithm, which captures any hidden preferences that are not obtained in the descriptive analytics stage. We then use prescriptive methods to generate the routes. Our approach allows us to blend the advantages of all facets of data science in a single collaborative framework, which is effective in learning the preferences and generating high-quality routes. A preliminary version of our descriptive method received the third-place award in the 2021 Amazon Last-Mile Routing Research Challenge.