01698028

Component

Taxi GPS trajectory data has been widely used in transportation research. However, the data is often corrupted with noises, of which the most commonly seen and consequential is mislabeled passenger occupancy status, which indicates whether there is a passenger in the taxi when a GPS point is recorded. This study develops an input-output hidden Markov model (IO-HMM) to identify and correct the mislabeled occupancy status in taxi trajectory data. Given the observed taxi moving behaviors and the recorded occupancy status, as well as specific spatiotemporal contexts, IO-HMM generates a series of predictions of the unknown taxi occupancy states. Since the true occupancy states are not available, the expectation-maximization (EM) algorithm is applied to train the model. The proposed model is experimented on a large taxi trajectory data set collected in Shenzhen, China. Results of numerical experiments show that the proposed model generates reliable predictions of taxi occupancy status, and consistently corrects the wrong labels in the data at a relatively high accuracy. Moreover, the proposed model shows reasonable robustness towards random errors manually added to ground truth trajectories.

This paper was sponsored by TRB committee ABJ70 Standing Committee on Artificial Intelligence and Advanced Computing Applications.

19-05030

English

7p

2019

Transportation Research Board 98th Annual Meeting

Digital/other

Maps; References (17) ; Tables

Data quality; Mathematical models; Passengers; Taxicab companies; Taxicabs; Vehicle occupancy; Vehicle trajectories

Shenzhen (China)

Data and Information Technology; Operations and Traffic Management; Passenger Transportation; Vehicles and Equipment

Transportation Research Board Annual Meeting 2019 Paper #19-05030

TRIS, TRB, ATRI

Dec 7 2018 9:44AM