Bi-objective feature selection in high-dimensional datasets using improved binary chimp optimization algorithm

• Published in: International journal of machine learning and cybernetics Vol. 15; no. 12; pp. 6107 - 6148
• Main Authors: Al-qudah, Nour Elhuda A., Abed-alguni, Bilal H., Barhoush, Malek
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.12.2024; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Artificial Intelligence; Complex Systems; Complexity; Computational Intelligence; Continuity (mathematics); Control; Datasets; Engineering; Feature selection; Heuristic methods; Integer programming; Machine learning; Mechatronics; Methods; Mutation; Optimization; Optimization algorithms; Optimization techniques; Original Article; Pattern Recognition; Performance enhancement; Robotics; Search methods; Statistical analysis; Statistical tests; Systems Biology; Transfer functions; High-dimensional datasets; Feature selection; Chimp optimization algorithm; Hill climbing algorithm; Opposition-based learning; Lévy flight
• ISSN: 1868-8071; 1868-808X
• DOI: 10.1007/s13042-024-02308-y

The machine learning process in high-dimensional datasets is far more complicated than in low-dimensional datasets. In high-dimensional datasets, Feature Selection (FS) is necessary to decrease the complexity of learning. However, FS in high-dimensional datasets is a complex process that requires the combination of several search techniques. The Chimp Optimization Algorithm, known as ChOA, is a new meta-heuristic method inspired by the chimps’ individual intellect and sexual incentive in cooperative hunting. It is basically employed in solving complex continuous optimization problems, while its binary version is frequently utilized in solving difficult binary optimization problems. Both versions of ChOA are subject to premature convergence and are incapable of effectively solving high-dimensional optimization problems. This paper proposes the Binary Improved ChOA Algorithm (BICHOA) for solving the bi-objective, high-dimensional FS problems (i.e., high-dimensional FS problems that aim to maximize the classifier’s accuracy and minimize the number of selected features from a dataset). BICHOA improves the performance of ChOA using four new exploration and exploitation techniques. First, it employs the opposition-based learning approach to initially create a population of diverse binary feasible solutions. Second, it incorporates the Lévy mutation function in the main probabilistic update function of ChOA to boost its searching and exploring capabilities. Third, it uses an iterative exploration technique based on an exploratory local search method called the β -hill climbing algorithm. Finally, it employs a new binary time-varying transfer function to calculate binary feasible solutions from the continuous feasible solutions generated by the update equations of the ChOA and β -hill climbing algorithms. BICHOA’s performance was assessed and compared against six machine learning classifiers, five integer programming methods, and nine efficient popular optimization algorithms using 25 real-world high-dimensional datasets from various domains. According to the overall experimental findings, BICHOA scored the highest accuracy, best objective value, and fewest selected features for each of the 25 real-world high-dimensional datasets. Besides, the reliability of the experimental findings was established using Friedman and Wilcoxon statistical tests.