A Novel Clustering-Based Hybrid Feature Selection Approach Using Ant Colony Optimization

Feature selection is an essential task in the field of machine learning, data mining, and pattern recognition, primarily, when we deal with a large number of features. Feature selection assists in enhancing prediction accuracy, reducing computation time, and creating more comprehensible models. In feature selection, each feature has two possibilities, either it would be taken for computation or not, which implies for n number of features, there are 2n\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2^{n}$$\end{document} possible feature subsets. So, identifying a relevant feature subset in a reasonable amount of time is an NP-hard problem, but by using an approximation algorithm, a near-optimal solution can be achieved. However, many of the feature selection algorithms use a sequential search strategy to select relevant features, which adds or removes features from the dataset sequentially and leads to trapped into a local optimum solution. In this paper, we propose a novel clustering-based hybrid feature selection approach using ant colony optimization that selects features randomly and measures the qualities of features by K-means clustering in terms of silhouette index and Laplacian score. The proposed feature selection approach allows random selection of features, which allows a better exploration of feature space and thus avoids the problem of being trapped in a local optimal solution, and generates a global optimal solution. The same is verified when compared with another state-of-the-art method.