Short-Term Traffic Flow Prediction Using Google Traffic Data

Abstract

The latest statistics show that by 2050, 60% of the world’s population will live in cities, which is expected to double the number of cars around 2.5 billion by 2050. Therefore, today and in the near future, traffic congestion has become an inevitable situation in the growing metropolitan areas of the world. Therefore there is a need for the development of a methodology of obtaining reliable and consistent traffic jams data to manage and come up with alternative routes to avoid traffic congestions which allow authorities to provide efficient service to society. This project carried out with the aim, to develop a methodology to integrate public data useful for short-term traffic flow prediction. The traffic data were extracted from capturing images of the Google map traffic layer in a particular area using Google Maps JavaScript API. All the predictions were achieved on experimenting only on the Sri Lankan contest where images are captured on a particular area of Colombo. These sequential images formed both the input and the prediction target are spatiotemporal sequences. Therefore a Convolutional LSTM (ConvLSTM) model approach was used for the traffic flow prediction, considering three main procedures which are based on predicting time 20 minutes, 30 minutes, and 1 hour in the future. The experiment that was carried 20 minutes into the future, has gained results in the precision of 0.551 and F1-measure of 0.647. This study yielded a profitable and affordable methodology for short-term traffic flow prediction appropriately for Sri Lankan Context.