Applying the Support Vector Machine to Predicting Headway-Based Bus Bunching

Bus bunching seriously damages the quality of transit service. This paper proposes a predictive approach to detect the occurrence of bus bunching depending on transit smart card data analysis. Several different support vector machine (SVM) algorithms are established to detect bus bunching based on the stop-level headway irregularity. An empirical experiment in Changzhou is conducted to analyze and compare the performances of different algorithms in practical applications. By comparing the prediction accuracy and computation efficiency, the optimal model is determined under different scenarios. The results suggest transit operators should use least squares support vector machine with linear kernel when the size of training dataset is large. Traditional SVM with linear kernel might be a better option for ensuring the prediction accuracy on smaller datasets. This strategy may improve transit service and increase transit ridership.