Kernel Credal Classification Rule – Application on Road Safety

Abstract

A credal partition based on belief functions has been proposed in the literature for data clustering. It allows the objects to belong; with different masses of belief; not only to the specific classes, but also to the sets of classes called meta-class which correspond to the disjunction of several specific classes. In this paper, a kernel version of the credal classification rule (CCR) is proposed to perform the classification in feature space of higher dimension. Each singleton class or meta-class is characterized by a center that can be obtained using many way. The kernels based approaches have become popular for several years to solve supervised or unsupervised learning problems. In this paper, our method is extended to the CCR. It is realized by replacing the inner product with an appropriate positive definite function, implicitly perform a nonlinear mapping of the input data into a high-dimensional feature space, and the corresponding algorithm is called kernel Credal Classification Rule( KCCR). We present in this work KCCR algorithm to powerful corresponding nonlinear form using Mercer kernels. The approach is applied for the classification of experimental data collected from a system called VehicleInfrastructure-Driver (VID), based on several representative trajectories observations made in a bend, to obtain adequate results with data experimentally realized based on the instructions given to drivers. The test on real experimental data shows the value of the exploratory analysis method of data. Another experiments using the generated and real data form benchmark database are presented to evaluate and compare the performance of the KCCR method with other classification approaches.