An Evolutionary Hybrid Feature Selection Approach for Biomedical Data Classification

Feature selection is an important function/subject in machine learning. It involves separating the relevant features of the data set and reducing its dimension by eliminating unnecessary data, leading to predictive performance. To this end, researchers use specific search methods to find an optimal subset of features. The aim of this study is developing a hybrid algorithm according to simulated annealing (SA) and grey wolf optimizer (GWO) to be applied in feature selection for biomedical data. Grey wolf algorithm optimizer is an innovative, bio-inspired method of optimization and, as the name suggests, reproduces the actual pattern of how grey wolves hunt in their natural habitat. Two feature selection methods (BGWO1-SA and BGWO2-SA) are presented here. For greater intensification of the suggested algorithm, the SA algorithm receives the inputs of the wolves' updated position in the last phase of both above-mentioned approaches. The proposed methods are compared with four competitors: particle swarm optimization, genetic algorithms, and two versions of GWO algorithm. The assessments were based on a set of challenging biomedical benchmarks, and the results showed that the presented methods outperform their rivals.