Feature Selection using Enhanced Nature Optimization Technique

Abstract

An essential study issue now is the preference of highly discriminative traits from a huge feature collection. By eliminating a significant number of noisy, redundant features, this has the potential to enhance classification performance while lowering the cost of system diagnostics. A feature selection process has been implemented using nature-inspired algorithms. Each of these algorithms needs its starting population to be initialized, and how well that initialization is done has a big impact on the outcome. This paper presents a newly hybrid nature-inspired Algorithm which is comprised by Harris-hawk Algorithm with Visual Geometry Group for selection of traits on High-Dimensional-datasets. Our main idea is to overcome the overfitting issue of feature selection and also overcome convergence problem arise in nature inspired algorithm by introducing visual geometry group Convolution neural network based deep neural network. Then, we compared our upgraded approach to the most significant nature-inspired optimization technique to show that our technique is more accurate and categorized using the Acute lymphoblastic leukemia & Breast cancer High Dimensional datasets.