Bus Travel Time Prediction Based on Relevance Vector Machine

Existing bus travel time prediction methods only provide a point prediction value of bus travel time. Relevance vector machine (RVM) is proposed for solving the problem. By using a probabilistic Bayesian learning framework, RVM can provide probabilistic prediction and obtain prediction value and variance of prediction error. For making use of historical data and current information, the running time of next segment of two days before, the running time of next segment of one day before, the latest running time of next segment of the same day, the running time of current segment of the same day and the dwell time of current stop of the same day are taken as five input variables in the model. And sample data are normalized for training and test. The example results show that the model has higher precision of prediction, and provide prediction interval of bus travel time which would be more valuable information for passengers.