V-shaped and S-shaped binary artificial protozoa optimizer (APO) algorithm for wrapper feature selection on biological data

This study addresses the limitation of feature selection (FS) problems in high-dimensional biomedical datasets. The high-dimensional datasets contain attributes that are deemed irrelevant, redundant, and noisy. Thus, the process of feature selection is a valuable initial step aimed at improving the performance of classification models through the identification and selection of a constrained set of significant and impactful features. Due to the NP-hard nature of feature selection, it is crucial to recognize that addressing these challenges requires the utilization of metaheuristic algorithms. However, since the feature selection problem is a discrete problem, the binary version of metaheuristic algorithms should be used. To overcome these challenges, this paper proposes a novel bAPO algorithm that leverages adaptive population dynamics for more efficient exploration and exploitation of the solution space. The proposed bAPO algorithm uses V-shaped and S-shaped transfer functions to obtain wrapper feature selection in biological data. There are eight different versions of the bAPO algorithm in this study that were evaluated with 14 well-known biological datasets. The obtained results have been analyzed with the fitness value, the number of selected features, k-nearest neighbors (KNN) accuracy, support vector machine (SVM) accuracy, and random forest (RF). Statistical validation using p-value analysis demonstrates the robustness and reliability of the results. The obtained findings suggest that the proposed bAPO algorithm provides a powerful method for tackling optimization problems, particularly in high-dimensional datasets. In fitness performance, the bAPO-V1 and bAPO-V2 (27.70%) demonstrate superior performance, and in terms of reduced features, the bAPO-V2 (36.36%) algorithm achieved good performance.