Random selection of samples and features for getting general accuracy of classifier combination

The goal of pattern classification is to classify class-unknown samples correctly, in which a classifier is trained from a finite number of samples. Multiple classifier si-stems are trials to overcome to a single (strong) classifier by combining some (weak) classifiers. In this paper, we propose a novel combining method in which each element classifier is constructed by a random selection of both training samples and features. Experimental results on two real datasets are shown. As a result, the performance of a single classifier was improved by combining them with this method.