Long-Term Individual Trip Pattern Prediction of Bus Passengers Using Smart Card Data: A Bayesian Method Based on Feature Selection

Abstract

Abstract Travel demands of bus passengers tend to be diversified and personalized, for targeted demand management and dynamic system operations, it is important to predict individual trip patterns. Smart card records can be combined with other data to infer the future trip patterns of passengers. This provides an effective method to study individual travel behavior in the field of public transportation. This paper proposes a methodology for predicting the travel patterns of individual bus passengers, including the prediction of trip start time and trip start station. Based on a passenger segmentation model, passengers with more frequent and regular bus trips are extracted first. Then combined with the information provided by date attributes and weather data, the most relevant influencing feature of the individual traveler's trip on the day are selected as the prediction feature. Afterwards, we use bayesian models to predict the individual traveler's future trip start time and station. The proposed methodology is tested using the long-term observation of smart card data from 1000 users of bus system in Kunshan, Jiangsu Province over one year. Our prediction achieves median accuracy levels over 65% and 70% for trip start time t and trip start station s. The proposed model can be easily applied to transport demand management. It can also be used for the design and optimization of customized public transport systems. In this way not only will bus passengers be better served, but the overall level of traffic control and management in the city will also be effectively improved.