A Comparison of Deep Learning Methods for Urban Traffic Forecasting using Floating Car Data

Cities today must address the challenge of sustainable mobility, and traffic state forecasting plays a key role in mitigating traffic congestion in urban areas. For example, predicting path travel time is a crucial issue in navigation and route planning applications. Furthermore, the pervasive penetration of information and communication technologies makes floating car data an important source of real-time data for intelligent transportation system applications. This paper deals with the problem of forecasting urban traffic when floating car data is available. A comparison of four deep learning methods is presented to demonstrate the capabilities of the neural network approaches (recurrent and/or convolutional) in solving the traffic forecasting problem in an urban context. Different tests are proposed in order to not only evaluate the developed deep learning models, but also to analyze how the penetration rates of floating cars affect forecasting accuracy. The presented experiments were designed according to a microscopic traffic simulation approach in order to emulate floating car data fleets, which provide vehicle position and speed, and to validate the obtained results. Finally, some conclusions and further research are presented.