Unsupervised feature ranking based on representation entropy

Feature ranking and selection play an important role in many areas of Machine learning. Most of the work found in the machine learning literature concerns itself with supervised dimensionality reduction where each instance of the dataset is attached with a class label. In this paper, we present an algorithm that ranks the features of an unlabeled dataset based on the concept of representation entropy. Entropy, in its different forms, has been successfully applied to the problem of feature ranking and selection. Representation entropy, used in this paper for the purpose of ranking features is based on the well known concept of principal components. The results obtained by the new algorithm are compared with the Relief-F, SUD algorithm and SVD-entropy based algorithm for various datasets and analyzed.