Evaluation of Two Methods for Identifying Trip Purpose in GPS-Based Household Travel Surveys

Abstract

Data needs for developing travel demand models have increased at the same time that household travel survey (HTS) participation rates have generally fallen over recent decades. GPS-assisted HTS methods are recognized today as the most promising direction in further enhancement of individual travel data collection. The principal advantage of the GPS-assisted survey technology is that a full stream of locations visited by the person is identified with a high level of spatial and temporal resolution, but automatic identification of trip purpose remains an issue that is difficult to solve. This paper evaluates the performance of two methods, choice modeling and decision tree analysis, that can be used to build models capable of identifying trip purpose. The developed methods assume that basic household- and person-level data, typically collected in an HTS, are available, as are supporting spatial data sets. The methods presented are then evaluated for a case study that employed data from the 2011 Atlanta Regional Commission HTS. The developed models produced encouraging results with overall accuracy greater than 70% across all purposes and around 90% for mandatory activities (i.e., work and school). The performance of the developed models was evaluated in terms of error rates by purpose category and the impact of ancillary spatial data. The paper concludes with a summary of the findings and recommendations for practitioners.

Developing an optimised activity type annotation method based on classification accuracy and entropy indices

“…activity start times, activity durations, land use information) (Gong et al 2014;Wolf, Guensler, and Bachman 2001;Lu, Zhu, and Zhang 2012;S. Lee and Hickman 2014;Shen and Stopher 2013;Montini et al 2014;Lu and Zhang 2015;Simas-Oliveira et al 2014;Lu, Zhu, and Zhang 2013;Feng and Timmermans 2015;Nurul Habib and Miller 2009). In the field of ITS (Intelligent Transportation Systems), transportation data mining methods that aim to integrate different data sources are generally denoted as data fusion (DF) techniques.…”

“…Many researches followed and advanced this domain of activity type annotation. Some focused on finding more and better data (Lu, Zhu, and Zhang 2012;Shen and Stopher 2013;Montini et al 2014;Lu and Zhang 2015;Simas-Oliveira et al 2014;Gong et al 2014), others experimented with different methodologies and data mining algorithms (S. Lee and Hickman 2014;Simas-Oliveira et al 2014;Lu, Zhu, and Zhang 2013). The activity type classes which are predicted (and the size of this set of classes) strongly impact the classification accuracy.…”

Modeling a Spatio-Temporal Individual Travel Behavior Using Geotagged Social Network Data: A Case Study of Greater Cincinnati

“…Comparing our study with some previous researches (Oliveira et al, 2014;Ermagun et al, 2017), our model seems to be more accurate. Our proposed approach can show useful information about the difference of the model predictions and the empirical pattern of destination choice.…”

“…Similarly, (Lu et al, 2013) explore the feasibility of automating trip purpose detection employing ML method with geospatial location data, the land use data, and GPS-based survey data. (Oliveira et al, 2014) used a two-level nested logit model (probabilistic) and a decision tree model (ML) to differentiate between 12 trip purposes. In their study, the decision tree model was more accurate and much faster to generate functioning models than nested logit model.…”

Comparative framework for activity-travel diary collection systems

“…All the potential trip purposes are classified into 13 categories and the reported overall accuracy is 43%. Oliveira et al [12] evaluate two methods for purpose inference that are based and choice modeling and decision tree analysis. These methods rely on GIS land use and points of interest datasets, and the reported overall accuracy is above 70% for 12 categories.…”