Feasible Adaptation Criteria for Hybrid Wavelet - Large Margin Classifiers

In the context of signal classification, this paper assembles and compares criteria to easily judge the discrimination quality of a set of feature vectors. The quality measures are based on the assumption that a Support Vector Machine is used for the final classification. Thus, the ultimate criterion is a large margin separating the two classes. We apply the criteria to control the feature extraction process for signal classification. Adaptive features related to the shape of the signals are extracted by wavelet filtering followed by a nonlinear map. To be able to test many features, the criteria are easily computable while still reliably predicting the classification performance. We also present a novel approach for computing the radius of a set of points in feature space. The radius, in relation to the margin, forms the most commonly used error bound for Support Vector Machines. For isotropic kernels, the problem of radius computation can be reduced to a common Support Vector Machine classification problem.