Bus Arrival Time Prediction And Bus Scheduling: A Combination Of Machine Learning And Large Neighborhood Search Approaches

Abstract

Bus system development has been paid more attention due to its convenience. To develop a bus system, Punctuality and regularity need to be improved. Good bus arrival time prediction models help improve punctuality while suitable bus scheduling enhance regularity. In this study, Machine Learning (ML) models were first built to predict bus arrival time at every bus stops. Then, the bus scheduling problem was solved using the Large Neighbourhood Search (LNS) framework. The result indicated that the arrival time of the current bus to a stop can be explained mostly by the travelling time to the same stop of preceding buses and its distance to the stop. Regarding the scheduling problem, the presented methodology was able to solve small to relatively large problems within a reasonable time. Future work should focus on adding more meaningful features to the ML models and improve the current LNS by modifying the degree of destructions and combining more destroy and repair methods