Data-driven Prediction of One-way Bus Running Time: An Integrated Model

Abstract

The one-way running time of the bus has been paid a lot of attention especially for passengers to decide their travels and for bus companies to make schedules. Recent research provide some machine learning algorithms for the prediction of similar topics, however without satisfactory results. In order to improve the predictive accuracy of bus one-way running time, four machine learning algorithms including LSTM, MLR, KNN, and XGBoost are selected and integrated through a linear regression model as an effectively integrated model. More specifically, dynamic and static factors of the one-way running time for public transportation vehicles are firstly analyzed and identified. In particular, weather and holidays are included for improving the effectiveness. Then the integrated model is applied and validated by using the operational data of Beijing Bus Line One. Finally, a comparative analysis is conducted on the results of using respectively singular methods and the integrated model which further validates the effectiveness of the integrated model.