2017 Volume 73 Issue 5 Pages I_517-I_526
Probe person survey has received great deal of attentions, as it can automatically collect traveler's spatiotemporal traces using sensors like GPS, making a long-term travel survey easier. However, it is still difficult to collect all of the survey results automatically, because trip purpose—one of the most important characteristics of travel—is not possible to be identified based on such sensor data itself, making a long-term survey costly. This study proposes trip purpose estimation method based on automated sensor data and minimal manual questionnaires, in order to reduce survey burden on respondent and enable a long-term survey. In the proposed method, trip purpose is estimated by a classifier with sequential learning. Specifically, if the estimation confidence is high enough, the estimation result is considered as a survey result. Otherwise, the method asks actual trip purpose to the respondent, and then manually answered trip purpose is used to update the classifier. As results, the method is expected to reduce survey burden on respondents by automatically estimating purposes of recurrent trips, while keep data quality by asking purposes of irregular trips. Empirical features of the proposed method is validated by emulating the method on actual survey data.