Evaluation of a NeuroFuzzy Unsupervised Feature Selection Approach

Abstract

Dimensionality reduction is a commonly used step in machine learning, especially when dealing with a high dimensional space of features. The original feature space is mapped onto a new, reduced dimensionality space and the examples to be used by machine learning algorithms are represented in that new space. The mapping is usually performed either by feature extraction or feature selection. Feature extraction involves constructing some new features from original feature set. Feature selection involves selecting a subset of the original features from original feature set without transformation. Feature selection can be implemented either by feature ranking or subset selection. Feature ranking is an approach in which all the features are ranked based on some criteria. In this project, Feature ranking algorithm has been implemented. Work presented here includes the implementation of UFSNF for ranking different features using the fuzzy evaluation index with neural networks. The results (ranks) obtained from UFSNF have been compared with the ranks obtained by Relief-F evaluator using four clustering techniques EM, k-Means, Farthest First and Hierarchical. For the experimental study, benchmark datasets from the UCI Machine Learning Repository have been used. From the study, it is found that the newly proposed algorithm, UFSNF in some cases exceeds the performance of Relief-F.