Enhanced coati optimization algorithm using elite opposition-based learning and adaptive search mechanism for feature selection

• Published in: International journal of machine learning and cybernetics Vol. 16; no. 1; pp. 361 - 394
• Main Authors: Qtaish, Amjad, Braik, Malik, Albashish, Dheeb, Alshammari, Mohammad T., Alreshidi, Abdulrahman, Alreshidi, Eissa Jaber
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.01.2025; Springer Nature B.V
• Subjects: Accuracy; Adaptive algorithms; Adaptive search techniques; Algorithms; Artificial Intelligence; Basic converters; Classification; Complex Systems; Computational Intelligence; Control; Convergence; Datasets; Engineering; Feature selection; Genetic algorithms; Heuristic; Machine learning; Mechatronics; Methods; Mutation; Optimization; Optimization algorithms; Original Article; Pattern Recognition; Robotics; Searching; Support vector machines; Systems Biology; Coati optimization algorithm; Feature selection; Opposition-based learning; Optimization
• ISSN: 1868-8071; 1868-808X
• DOI: 10.1007/s13042-024-02222-3

The rapid rise in volume and feature dimensions is negatively impacting machine learning and many other areas, leading to worse classification accuracy and higher computational costs. Feature Selection (FS) methods are crucial to lessen feature dimensionality, which act by removing attributes like irrelevant and less informative information which may have a detrimental impact on the performance of classifiers. This paper presents an Enhanced variant of the Coati Optimization Algorithm (ECOA) that features a better search ability than the basic COA. The COA algorithm was newly evolved to imitate the behavior of coatis when they hunt and attack iguanas as well as when they try to flee from predators. Although the authors of this algorithm state that it is promising, it occasionally exhibits poor search performance and early convergence. To mitigate these issues, the ECOA algorithm was proposed that makes use of elite opposite-based learning in addition to some adaptive search mechanisms. ECOA is expected to have an improved search mechanism and can prevent trapping at local optimum, depending on the mutation, mutation neighborhood search, and rollback procedures. Moreover, it enhances population variety and convergence rate. The COA and ECOA algorithms were used to solve FS problems by selecting optimal feature subsets based on a binary version of each adopted algorithm and the k-Nearest Neighbor (k-NN) classifier. To assess the performance of the Binary ECOA (BECOA), a number of experiments was performed on 24 datasets collected from many sources. Further, six criteria-sensitivity, specificity, classification accuracy, fitness value, number of chosen features, and run time-were used to assess the performance of BECOA. Experimental findings show the excellence of BECOA over other k-NN based FS methods, including Binary COA (BCOA) and other binary optimization methods, in a number of assessment aspects. In particular, among the 24 datasets deemed, BECOA, which yielded the best overall results among all other competing binary algorithms, was able to exclusively outperform the others in 7 datasets in terms of classification accuracy, 11 datasets in terms of specificity, 5 datasets in terms of sensitivity, 10 datasets in terms of number of selected features, 4 in terms of run-time, and 14 datasets in terms of fitness values.