Short time traffic speed prediction using pattern recognition and feature selection methods

In this study we predict traffic speed on Istanbul roads using RTMS (remote traffic microwave sensor) speed measurements obtained from the Istanbul Municipality web site. We use two different pattern recognition methods, k-nearest neighbor (kNN) and support vector regression machine (SVM). In order to predict the speed at a short time (5 minutes to 60 minutes) ahead, we use speed measurements taken at different time intervals before the prediction. We find out that both long term (1 day or 1 week) speed information and short term (5 minutes to 60 minutes) speed information play an important role for short time speed prediction. We use backward feature selection algorithm to find out the most important features for speed prediction. In addition to speed measurements for the same sensor at different times, we use speed measurements from different sensors to predict speed at a certain sensor point. We find out that using additional sensors results in better speed prediction. We also find out that using SVM results in better prediction than KNN for this problem.