Automatic prediction of icy conditions on roads using a LS-SVM classifier

Authors:	Beguin Sophie, UCL, Belgium Bastin Georges, UCL, Belgium Wertz Vincent, UCL, Belgium
Topic:	7.4 Transportation Systems
Session:	Transportation Systems I
Keywords:	LS-SVM, ice prediction

Abstract

This paper describes a method for predicting the presence orabsence of ice on the road. The method is based on a LeastSquares Support Vector Machine applied to data from the road inWallonia (Belgium). It is shown that including a prediction of theair temperature given by a meteorological center in the modelhelps having better accuracy. In this application, 95\% accuracyhave been achieved for a 3 hours prediction horizon, and 92\% for6, 12 and 24 hours horizon.