Relief wrapper based Kernel Partial Least Squares subspace selection

Kernel partial least squares method can obtain nonlinear novel features for further classification and other tasks, the dimension of extracted kernel space is usually very high, there still may contain irrelevant and redundant features, so using feature selection to select the most discriminative and informative features for classification or data analysis is important, but there are few attentions to it until now. Here we propose a novel method which firstly uses kernel partial least squares as a nonlinear feature extraction method to get a basis set, and then uses the relief wrapper, one of the hybrid feature selection algorithms, to select the most discriminative features. The selected features form a subspace of the kernel space, where different state-of-the-art classification algorithms can be applied for classification. Experimental results on three microarray datasets validate the efficiency and accuracy of our method.