A trip purpose inference framework using spatial clustering and bayesian probability

Abstract

Taxis are one of the most widely used modes of transport among urban communities. The use of GPS devices in modern taxi vehicles has enabled the estimation of travel patterns through emitted and collected massive scale trip records. The only necessity that requires for this is a suitable trip purpose inference model as the GPS data are unable to provide the exact purpose of a trip but the neighborhood of travelers’ destination. Thus, this study attempted to develop a trip purposes inference framework that can be used reliably in uncovering travel patterns. The proposed framework consists of three layers: (1) Trip purpose imputation for regular trips using spatial clustering, (2) Identifying the trips attracted to residential trips, and (3) Purpose inference using Bayesian probability. The model was tested using taxi trips data from a service provider operating in Colombo District, Sri Lanka, and compared that with the activity proportions data taken from a household travel survey. The results indicates that the proposed model is capable of providing plausible travel patterns through identified spatial dynamics and temporal patterns.