Analysis of Evolutionary Intelligence Approaches for Feature Selection and Open Challenges

The superiority of stochastic approaches for feature selection algorithms has been demonstrated over the past two decades, and tens of new ones are developed and tested annually on UCI datasets. It can provide a way to reduce processing costs, improve forecast accuracy, and improve understanding of data structures. Evolutionary Intelligence (EI) algorithms are more promising approaches to feature selection problems because stochastic methods are among the many feature selection strategies that have many local solutions. However, the work is challenging due to two factors: feature communication and a large solution space. In this study, using a wrapper-based framework, we focus on the basic algorithmic structures of evolutionary intelligence for feature selection in order to choose the best feature subset for classification. This study examined the classification accuracy and convergence speed of five Evolutionary Intelligence feature selection methods across the five UCI datasets. Particle Swarm Optimisation (PSO), Harmony Search (HS), Rhinoceros Search Algorithm (RSA), Genetic Algorithm (GA), and Gravitational Search Algorithm (GSA) are the five EI algorithms that are utilised in their most basic form. Recent algorithms, such as HSA and GA, are used to identify the best feature subsets in the absence of empirical data. Five benchmark datasets are taken from the UCI repository in order to verify the outcomes of the chosen EIs.