A systematic review on the use of machine learning in last mile delivery vehicle routing problems

Vehicle Routing Problem has been a classic optimization deadlock, which has been present in operational research for the last 50 years. It aims to find solutions to minimize distances to be covered in delivery demands. Over time and considering the development of logistics, some variants have emerged, such as Last Mile Delivery. These comprise the final stretch, from the regional distribution center to the end user. In this article we have shown a systematic review of the literature on the Last Mile Delivery Vehicle Routing Problem, aiming to find Machine Learning methods and techniques used for optimization. The research is focused on the analysis of scientific publications made between 2016 and 2021 in the following databases: ACM Digital Library, IEEE Xplorer, Science Direct, Springer Link, Wiley Online Library and Brazilian Digital Library of Theses and Dissertations. We obtained a sample of 103 papers, which were categorized according to the solutions presented and the methods applied. The results indicate that metaheuristics and hybrid or mixed methods are the most used. The percentage of participation of Machine Learning is not significant compared to traditional methods, being present in only 6% of publications.