Two-Stage Data-Driven Trip Purposes and Socio-Economic Attribute Inference Framework of Public Transit Riders
Chao Yang, Wentao Dong, Chengcheng Yu, Y ZHANG, Yingfei Tu
Data-driven research is becoming a new paradigm in transportation, but the natural lack of individual socio-economic attributes in transportation data makes research such as activity purpose inference and mobility pattern identification lack convincingness and verifiability. In this paper, a two-stage trip purpose and socio-economic attributes inference model is proposed based on travel resident survey and smart card data. In the first stage, the trip purpose of each trip is inferred by a combination of rule-based and XGBoost models. In the second stage, based on the trip purpose, a machine-learning model is built to infer the socio-economic attributes of individuals. A teacher–student model based on self-training is then applied to the models above to transfer them to smart card data. The impact of independent variables of the socio-economic attributes inference model is also investigated. The results show that the proposed framework achieves an overall accuracy of 92.7% for trip-purpose inference, and an average accuracy of 76.3% across the three socio-economic attributes (age, job status, and income). Travel time, arrival time, departure time, and purpose of the first two trips are the most important factors on age and job status, while the land price related to jobs and housing is significant when inferring individuals’ incomes.