Real-time classification of aggregated traffic conditions using relevance vector machines

This paper examines the theory and application of a recently developed machine learning technique namely Relevance Vector Machines (RVMs) in the task of traffic conditions classification. Traffic conditions are labelled as dangerous (i.e. probably leading to a collision) and safe (i.e. a normal driving) based on 15-minute measurements of average speed and volume. Two different RVM algorithms are trained with two real-world datasets and validated with one real-world dataset describing traffic conditions of a motorway and two A-class roads in the UK. The performance of these classifiers is compared to the popular and successfully applied technique of Support vector machines (SVMs). The main findings indicate that RVMs could successfully be employed in real-time classification of traffic conditions. They rely on a fewer number of decision vectors, their training time could be reduced to the level of seconds and their classification rates are similar to those of SVMs. However, RVM algorithms with a larger training dataset consisting of highly disaggregated traffic data, as well as the incorporation of other traffic or network variables so as to better describe traffic dynamics, may lead to higher classification accuracy than the one presented in this paper.