01622573

Component

In today’s mobile world, long-distance journeys are responsible for almost half of overall traffic. Traditionally, surveys have been used to gather data needed for the analysis of travel demand. Due to the high response burden and memory issues, respondents are known to underreport the number of journeys. Thus, alternative data sources are becoming more important. These sources collect the data passively, e.g. using GPS or GSM networks. The limitation of passively collected data is the lack of semantic information, especially trip purposes. Additionally, socio-demographic information is missing making it difficult to impute the purpose. This paper shows how one can predict the tour purpose without the need of socio-demographic information. The solution extends the well known random forest approach. Instead of a single random forest, a series of random forests is applied to classify the data. Attributes of the tours are used in order to overcome the lack of personal information. The presented approach is applied to a long-distance tour data set based on mobile phone billing data covering 5 months of mobile phone usage in France. The training set for the algorithm was taken from a national travel survey covering the same range. The purpose classification algorithm provides shares of trip purposes that are comparable to the shares in the used survey indicating that the approach is valuable.

This paper was sponsored by TRB committee ABJ10 Standing Committee on National Transportation Data Requirements and Programs.

01618707

17-02426

English

17p

2017

Transportation Research Board 96th Annual Meeting

Digital/other

Figures; References; Tables

Algorithms; Classification; Decision trees; Smartphones; Travel surveys; Trip purpose

France

Data and Information Technology; Highways; Planning and Forecasting

Transportation Research Board Annual Meeting 2017 Paper #17-02426

TRIS, TRB, ATRI

Dec 8 2016 10:54AM