Bus Journey and Arrival Time Prediction based on Archived AVL/GPS data using Machine Learning

With a surge in the number of vehicles, urban traffic congestion has increased in recent years. This has led to increased travel times and decreased accessibility and mobility. One option to mitigate this issue is to promote the use of public transport, including buses. To encourage the use of buses, there is a need to provide reliable travel time and arrival information to commuters.In this study, we propose and develop predictive models to predict bus journey and arrival times based on historical AVL/GPS data, bus route and bus stop information. There were two parts to this study. The first was to predict overall journey times and the second was to predict bus arrival times at bus stops.To estimate total bus journey times, three models were developed using Linear Regression, Artificial Neural Network (ANN) and Long Short Term Memory Network (LSTM). Evaluation on a ground-truth dataset shows that LSTM outperformed the Linear Regression model and its performance was comparable to that of ANN. To predict bus arrival times at bus stops, three different models, namely Historical Averaging, Linear Regression and Gradient Boosting are proposed. Experimental results show that Gradient Boosting outperformed the other models and is more robust in predicting arrival times.Our study supports the idea that it is possible to predict bus journey time with reasonable accuracy using historical GPS observations and bus route information only.