A new feature selection approach with binary exponential henry gas solubility optimization and hybrid data transformation methods.

In the common classification practices, feature selection is an important aspect that highly impacts the computation efficacy of the model, while implementing complex computer vision tasks. The metaheuristic optimization algorithms gain popularity to obtain optimal feature subset. However, the feature selection using metaheuristics suffers from two common stability problems, namely premature convergence and slow convergence rate. Therefore, to handle the stability problems, this paper presents a fused dataset transformation approach by joining weighted Principal Component Analysis and Fast Independent Component Analysis Techniques. The presented method solves the stability issues by first transforming the original dataset, thereafter newly proposed variant of Henry Gas Solubility Optimization is employed for obtaining a new feature's subset. The proposed method has been compared with other metaheuristic approaches across seven benchmark datasets and observed that it selects better features set which improves the accuracy and computational complexity of the model.