Feature Selection Based on Fuzzy Mutual Information
Abstract: In the framework of fuzzy rule-based models for regression problems, we propose a novel approach to feature selection based on the minimal-redundancy-maximal-relevance criterion. The relevance of a feature is measured in terms of a novel definition of fuzzy mutual information between the feature and the output variable. The redundancy is computed as the average fuzzy mutual information between the feature and the just selected features. The approach results to be particularly suitable for selecting features before designing fuzzy rule-based systems (FRBSs). We tested our approach on twelve regression problems using Mamdani FRBSs built by applying the Wang and Mendel algorithm. We show that our approach is particularly effective in selecting features by comparing the mean square errors achieved by the Mamdani FRBSs generated using the features selected by a state of the art feature selection algorithm and by our approach.
Loading