Applying Artificial Neural Network For Long-Term Road Travel Time Prediction

This thesis focuses on the prediction of long-term travel time on routes with a data-driven approach.A methodology using an artificial neural network called multiple layer percep-tron has been implemented in order to forecast the travel time time series.6-year travel time data(2011-2016)of the road 4638 in California(USA)were used with a timestep of 5 minutes to test the methodology.First,several experiments were conducted in order to find the best hyper-parameters of the network.Then,additional features like the weather forecast,the number of closed lanes,the luminosity and the holiday calendar have been added to the network input dataset in order to test if they increase the accuracy of the model.Finally,a re-lationship between the weekly average travel time variance of the travel time dataset and the prediction error has been observed.A method has been implemented in order to decrease the impact of the variance of the travel time dataset on the prediction error and gave good results.