Abstract:

To collect passenger flow information of public transit from the widely applied smart card fare payment systems, a new method is proposed to infer the stops at which passengers holding smart cards board the bus from smart card fare data and bus schedules. The method first classifies two sequential swipes to decide whether they occur at the same stop with naive Bayes classifier. Travel times are then estimated from the naive Bayes classifier results using maximum likelihood estimation, dynamic programming and quadratic programming methods. To solve the problem with imprecise initial parameters, a coordinate descent method is applied. It updates parameters and estimates values alternatively until convergence. An experiment is designed to test this algorithm with realworld data, and it proves that the error of this method is small and the convergence is fast.

Key words: public transit system, public transit smart card, naive Bayes classifier, dynamic programming, quadratic programming, coordinate descent method