Entropy Dependency-based Unsupervised Feature Selection

Machine-learning techniques often face significant challenges with high-dimensional data due to extensive memory requirements and higher processing times. In this paper, a novel unsupervised feature selection method called Entropy Dependency-based Un-supervised Feature Selection (EDUFS) is proposed. This method aims to select an appropriate and concise subset of features while minimizing redundancy by evaluating and considering entropy of each feature during optimization. EDUFS employs entropy, an information theory concept, to measure the correlation between features. By calculating the entropy for each feature and eliminating redundant ones, our method reduces the feature set while preserving essential information in the form of Laplacian from the original dataset. Experimental evaluations across six datasets and seven different entropy measures demonstrate that our proposed approach is efficient and outperforms state-of-the-art unsupervised feature selection methods in most cases.