A neural network approach for predicting speeds on road networks

Official URL: http://dx.doi.org/10.1109/SIU.2018.8404705

It is possible for routing and navigation applications to provide more accurate and more effective route planning solutions by accurately predicting the traffic density or vehicle speed. Numerous methods and approaches have been studied to achieve this objective; however, they have mainly focused on the short-term traffic prediction. In addition, the studies that attempt to provide mid-and long-term predictions tend to show unacceptable accuracy levels. In this study, we employ Artificial Neural Networks (ANN). They will combine the predictions made by various time series forecasting methods to make mid-and long-term speed predictions. In the experimental study, we utilize floating car speed data on two routes collected by GPS devices with 1-minute intervals over a five month-period. The results reveal the superior performance of ANN and show that it provides accurate predictions over a 30-minute time interval.