A New Fast Filter-based Unsupervised Feature Selection Algorithm Using Cumulative and Shannon Entropy

Abstract

The feature selection process is indispensable for the machine learning area to avoid the curse of dimensionality. Hereof, the feature selection techniques endeavor to handle this issue. Yet, the feature selection techniques hold several weaknesses: (i) the efficacy of the machine learning methods could be quite different on the chosen features (ii) by depending on the selected subset, substantial differences in the effectiveness of the machine learning algorithms could also be monitored (iii) the feature selection algorithms can consume much time on massive data. In this work, to address the issues above, we suggest a new and quick unsupervised feature selection procedure, which is based on a filter and univariate technique. The offered approach together regards both the Shannon entropy computed by the symmetry of the distribution and the cumulative entropy of the distribution. As a consequence of comparisons done with some cutting-edge feature selection strategies, the empirical results indicate that the presented algorithm solves these problems in a better way than other methods.