Random Subspace Method with Feature Subsets Selected by a Fuzzy Class Separability Index

Classifier combining techniques have become popular for improving weak classifiers in recent years. The random subspace method (RSM) is an efficient classifier combining technique that can improve classification performance of weak classifiers for the small sample size (SSS) problems. In RSM, the feature subsets are randomly selected and the resulting datasets are used to train classifiers. However, these randomly selected feature subsets give no guarantee of carrying the necessary discriminant information. Therefore, the performance of RSM is sometimes not so stable. A common intuition suggests that the employed classifiers in the ensemble should be as accurate as possible. In other words, if we employ techniques to ensure the performance of base classifiers, we then can make sure of the performance of RSM as well. In this study, a fuzzy class separability index is developed and introduced to RSM for finding more informative feature subsets and thus constructing more accurate classifiers. The experimental results on real datasets demonstrate that the proposed framework indeed yields better results than the original RSM, particularly in SSS cases.