Effectively Finding the Optimal Wavelet for Hybrid Wavelet - Large Margin Signal Classification

For hybrid wavelet – large margin classifiers, adapting the wavelet may significantly improve the classification performance. We propose to select the wavelet with respect to a large margin classifier and data to improve class separability and minimise the generalisation error. In this paper, we show that this wavelet adaptation problem can be formulated as an optimisation problem with polynomial objective function and investigate some techniques to solve it. In particular, we propose an adaptive grid search algorithm that efficiently solves the problem compared with standard optimisation techniques.