Ensemble Learning Prediction of Bus Travel Time Based on Big AVL Data

Unreliable bus transit affects both providers and users in many ways. With the widespread use of the automated vehicle location (AVL) system, rational utilization of AVL data to analyze bus service and predict bus travel time will play a critical role in regulating the entire public transport system and attracting more passengers. In this paper, we conducted a systematic analysis of the reliability of the bus travel time. The bus travel time shows a bimodal distribution and fits a lognormal distribution. Moreover, the variation in bus travel time is larger on weekdays than on weekends. Then, ensemble-based prediction models were used to predict travel time based on relevant findings. This study has promising implications for understanding bus travel time and offers a solution to rationalize bus scheduling for bus authorities.