Inferring socioeconomic attributes of public transit passengers using classifiers

Emerging new datasets in the public transit network, such as smart card datasets, are changing the transport research area to a data-rich one. While these datasets include a large number of passenger trips in the network, they miss the socioeconomic attributes of passengers. Estimating the socioeconomic attributes from trip attributes is a way to enrich these datasets. To do so, Household Travel Survey (HTS) can be used to learn the relation between trips and socioeconomic attributes of passengers. This paper compares the performance of three well-known classifiers of Naïve Bayes, Random Forest, and Support Vector Machine for estimating age and income of passengers in the public transit network using the HTS. The explanatory variables are considered as the start time of the trip, activity duration (time period between two consecutive trips), land use around the origin, and land use around the destination of trips. The target variables are age and income of passengers. HTS data, which include both explanatory and target variables, are used to train and validate the classifiers. Three measures including Accuracy, F-score, and Informedness are used to evaluate the performance of the classifiers. Also, the three classifiers are compared to a random classifier. The case study is the HTS from the South East Queensland (SEQ) between 2009 and 2012. Results show that the Naïve Bayes classifier is a better classifier for estimating the age and income of passengers from trip attributes