Link Duration Prediction in VANETs via AdaBoost

Abstract

In this paper, we present a link duration prediction method in VANETs. It utilizes AdaBoost algorithm to combine several link metrics, such as distance, difference in velocities, link lifetime, and etc., to form a predictor with a higher accuracy. The proposed method is applicable to different traffic scenarios and does not rely on any assumption of vehicles' velocity distribution. We evaluate the performance of this Adaboost-based link duration prediction algorithm by various traffic traces generated by SUMO that represents different typical scenarios. The evaluation result shows that, the proposed method effectively improve link prediction accuracy. Compared with other machine learning based solutions such as linear regression and support vector regression, the proposed method also shows less prediction error consistently.