Recursive decomposition probability model for demand estimation of street-hailing taxis utilizing GPS trajectory data

• Published in: Transportation research. Part B: methodological 2023-01, Vol.167, p.171-195
• Main Authors: Wang, Jianbiao, Miwa, Tomio, Morikawa, Takayuki
• Format: Article
• Language: English
• Subjects: Probabilistic model; Recursive decomposition; Street-hailing taxi; Taxi demand estimation
• ISSN: 0191-2615; 1879-2367
• DOI: 10.1016/j.trb.2022.11.014

•An analytic model is proposed to infer the real demand of street-hailing taxis utilizing GPS trajectory data.•The proposed model is validated through simulation analyses.•A case study utilizing real data is conducted for empirical analysis. The flexible and personalized street-hailing taxi service constitutes an indispensable component of urban mobility. However, most studies have focused only on the observed demand (pickup record) while ignoring the unmet demand. If based only on such analysis, the effectiveness of demand management policies and taxi searching strategies will be undermined. Motivated by this, we develop a recursive decomposition probability (RDP) model to efficiently estimate the demand (the summation of observed and unmet demands) of street-hailing taxis based on GPS trajectory data. In detail, we first partition the chronologically ordered passing records of taxis in individual road segments into several observations. Then, we utilize the proposed RDP model to derive the probability of each observation. Finally, the likelihood function is established, and the demand parameters for each road are estimated by maximum likelihood estimation. A simulation analysis is conducted to validate the performance of the proposed RDP model. It is shown that the demand estimated by the proposed model is much closer to the set demand than the observed demand under all testing scenarios. In addition, by comparing with benchmarks, the proposed RDP model needs no assumptions about the taxi arrival process and the numerical results show that it performs well. Finally, empirical studies are conducted with field data, the demand during the morning peak hour in 626 roads in Xi'an, China, is estimated by the proposed RDP model and the extended geographically weighted RDP model. The result indicates that the demand is obviously higher than the observed demand, emphasizing the importance of estimating demand in demand management.