The bus arrival time prediction using LSTM neural network and location analysis

Abstract

The accurate bus arrival time information is crucial to passengers for reducing waiting times at the bus stop and improve the attractiveness of public transport. GPS-equipped buses can be considered as mobile sensors showing traffic flows on road surfaces. In this paper, we present an approach that predicts bus arrival time using historical bus GPS information and real-time situation on the road. In this study, we divide bus arrival time into bus dwelling time at bus stops and bus travel time between stations and predict each of them separately. The clustering approach used to predict the travel time between stations, and then for each cluster, we apply LSTM NN to predict walking time between stations. The latency at each bus stop we evaluate by historical dwelling time and using location analysis to find the importance of the bus stop as a point of interest during prediction time. The study is trained and tested on GPS data collected from 1200 buses in a period of 3 months. According to tests results our method show small mean absolute error for buses that not far from departure station. The outcomes of this work can be used as an additional information for bus passengers to know possible bus coming time and to estimate possible travel time in bus journey. The method for arrival time prediction proposed in this research has several advantages. It considers historical bus travel time information, real time information, bus dwelling time, riding time, traffic lights and city facilities.