A Unified Framework for Predicting KPIs of On-Demand Transport Services

Abstract

Having a better understanding of the key performance indicators (KPIs, e.g., demand and unmet demand) in the next time slot (e.g., next hour) is important for on-demand transport services, such as Uber and DiDi, to improve the service quality. In addition to the spatio-temporal dynamics, KPIs of on-demand transport services are also affected by many exogenous factors from different domains, e.g., the traffic condition from transportation domain and the weather condition from meteorology domain. Therefore, this paper proposes a unified framework to fuse the data collected from different domains to predict multiple KPIs for on-demand transport services. As demonstrated by the experiments, the proposed framework can capture both long-term regularity and short-term dynamics, thus achieving a better performance than the existing solutions in predicting KPIs.