Data-Driven Policies for the Online Ride-Hailing Problem with Fairness

Ride-hailing is a prevalent transportation service that facilitates mobility in urban areas. A ride-hailing service system encompasses several research problems, including the operational assignment of on-demand ride requests to vehicles in real time. The literature suggests various approaches to address similar systems, mainly optimizing the system efficiency, but recent studies pointed out that these systems are likely to cause geographical unfairness among passengers. Such unfairness may imply, for example, that requests whose origin or destination is far from centralized locations may suffer from excessive service rejections. In this paper, we suggest a data-driven approach to design an online assignment policy to overcome this phenomenon. We formulate the online ride-hailing problem with fairness that seeks to maximize both efficiency and geographical fairness in the system while achieving an adequate balance between them. To solve this problem, we offer a new general method to develop online assignment policies based on solutions for offline versions of the problem. The new method suggests extracting information from these solutions to guide real-time assignment decisions, which are chosen using a data-driven algorithm. With a simulation study, we examine the performance of our online policies relative to dispatching rules using synthetic random data that represent a real city layout and movement. Some of these rules are commonly used in practice, and some are more sophisticated ones. Our results demonstrate the viability of our approach to designing online policies. Compared with other dispatching rules, the experiments show that the generated policies maintain a better trade-off between efficiency and geographical fairness and preserve stable performance regardless of the instance size in different system settings.