Mining travel time from smart card fare data

The wide applications of smart card payment techniques in public transit systems provide a new way of collecting travel time information. In this paper, one method is proposed to estimate travel times with smart card fare payment data and bus schedule data. The proposed method first classifies two sequential card swipes to infer if they occur at the same stop with Naive Bayesian Classifier (NBC). Travel time is estimated from the NBC results using Maximum Likelihood Estimation (MLE), Dynamic Programming (DP) and Quadratic Programming (QP) methods. In order to solve the problem with imprecise initial parameters, coordinate descent method is applied, which updates parameters and estimate values alternatively until it converges. An experiment with real-world data is designed to quantify the reliability of this algorithm and the outcomes is contrast with GPS data. It shows that the error of this method is small and the convergence is fast.