Vehicle collision prediction based on support vector data description

Abstract

At present, most researchers consider that the number of types of collision factors are limited when researching vehicle collision prediction. In addition, most studies are based on two-class or multi-class methods to study and predict vehicle collisions, but few researchers consider applying single-class classification methods to predict vehicle collisions. In this paper, the least squares method is used to select the 12 factors that affect the collision, and the single category classification method is used to predict the vehicle collision. SVDD (Support Vector Domain Description) was used as a model method for the experiments in this article, the factors (such as weather, traffic flow) that affect vehicle collision are used as the input of the model to train a minimum hypersphere. The validation sample set is validated, and the sample points falling within the sphere are the support vectors trained by the model, that is to say, if the (New) samples similar to the training sample set belong to the same class, those falling outside the sphere do not belong to the training sample set. In this paper, one class SVM is used as a comparative experiment. Finally, the experiment shows that SVDD obtains better prediction results, and the accuracy is better than one class SYTI model.