Short-term speed prediction on urban highways by ensemble learning with feature subset selection

Abstract

Accurate traffic speed prediction is essential in the development of intelligent transportation systems. Even though a lot of methods have been proposed for traffic prediction, few works pay attention to the application of ensemble learning and feature subset selection. In this paper, we propose an implementation of ensemble learning using combination of M5 model tree and bagging to tackle traffic speed prediction. A method to select optimal neighboring links as features for our prediction model is also introduced, and different feature subset selection methods are compared. Experimental results show that the proposed ensemble with feature subset selection outperforms both single model and nonparametric model (k-NN). © 2014 Springer-Verlag Berlin Heidelberg.