Bus Travel Times Prediction Using ARIMA Models

Buses play an important role in transporting people and goods in cities. It is also viewed as a promising way to alleviate urban congestion. One of the complaints about taking buses is that travelers do not know when the next bus would arrive, which causes anxiety among travelers. To make bus transportation more convenient, cities start to install GPS devices on the buses to track their locations in real-time, so as to provide predicted bus arrival times to travelers. In this way, travelers can better plan their trips and the ridership of buses can be promoted.The goal of this research is to investigate models that can predict bus arrival times using historical GPS locations of the buses. Specifically, we want to test their performance using real data from the city of Hefei. We focus on the ARIMA time series model, which has been shown as a popular model for this purpose. We conduct systematic experiments on the parameters of the ARIMA model. We find that the distance between successive stops and the width of time bins have noticeable effect on the accuracy of the prediction. In particular, when the width of time bins and/or the distance between successive stops increases, more accurate predictions can be obtained.We also benchmark our ARIMA model against predictions obtained using historical averages. We find that our models generate more accurate results.