Kernel-based classification of tissues using feature weightings

Abstract

In high-dimensional spaces classification methods could be more effective using various feature selection methods. The training procedure could be speeded up by decreasing the dimension of the feature space, and the classification method could be improved by removing noisy or irrelevant features. In this paper we present a new method which weights the features according to their importance instead of removing the negligible ones via kernel functions. It could be applied to a range of real-world problems. We tested it on several biological datasets like a small part of the UCI Learning Repository and SCOP and the Leukaemia AML-ALL databases, and obtained a significantly better classification performance than that using the usual unweighted method. © 2006, Penkala Bt.