A bus running length prediction method based on Gradient Boosting

Abstract

At present,China’s public transport companies rely on experienced staff to estimate the return time of vehicles and then conduct vehicle dispatch.This method often results in large errors and wrang decisions due to the lack of auxiliary prediction methods.Based on the actual needs of bus companies,a prediction method R-GBDT based on dynamic feature selection was proposed.The R-GBDT utilizes feature selection components and model parameter adjustment components to provide predictive components with feature combinations and parameters that conform to the line characteristics,then the fusion component combines the results of other components to form a framework for predicting the final time interval.The experimental results from real bus to off-site data show that compared with other algorithms,the method can greatly improve the accuracy of bus transit time prediction.

Keywords

• bus dispatch;arrival prediction;GBDT