Optimization-based Predictive Approach for On-Demand Transportation

Optimizing the use of vehicles is an essential task for sustainable and effective mobility-on-demand services. In a service, a driver aims to accept maximum customers, while a customer wants to minimize his/her waiting time before getting notifications/served. A service platform always faces a trade-off between the two stakeholders and their key performance indicators (KPIs), i.e., the number of accepted customers and waiting times. This paper addresses the problem of maintaining the best possible KPIs by optimizing the use of facilities with solving Dial-a-Ride problems (DARP). We propose a new framework named FORE-SEAQER (FORecast Enhanced StepwisE Allocator with Quick answER), which predicts whether incoming customers can ride in assigned cars using both real and predicted future requests, and decides whether the platform accepts requests as soon as possible. We experimentally evaluate our framework on real-world service log data from Japan and confirm that the proposed framework reasonably works.