A new multi-objective binary Harris Hawks optimization for gene selection in microarray data

• Published in: Journal of ambient intelligence and humanized computing Vol. 14; no. 4; pp. 3157 - 3176
• Main Authors: Dabba, Ali, Tari, Abdelkamel, Meftali, Samy
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.04.2023; Springer Nature B.V; Springer
• Subjects: Accuracy; Algorithms; Artificial Intelligence; Benchmarks; Bioinformatics; Cancer; Classification; Classifiers; Computational Intelligence; Computer Science; Engineering; Feature selection; Gene expression; Genes; Genetic algorithms; Methods; Multiple objective analysis; Mutation; Optimization; Optimization algorithms; Optimization techniques; Original Research; Robotics and Automation; Support vector machines; User Interfaces and Human Computer Interaction; Feature selection; Harris Hawks optimization algorithm; Swarm intelligence; Gene expression; Cancer classification; Microarray data
• ISSN: 1868-5137; 1868-5145
• DOI: 10.1007/s12652-021-03441-0

Cancer classification is one of the main applications of gene expression data (microarray data) and is essential for a comprehensive diagnosis of cancer treatment. Therefore, bio-inspired algorithms have developed several effective applications in the analysis of gene selection, which are one of the most effective applied in this domain. Harris Hawks optimization is a novel and recent algorithm that has an excellent balance between exploration and exploitation. This paper presents the first study on multi-objective binary Harris Hawks optimization (MOBHHO) for gene selection. We define gene selection as a problem, including two main conflicting objectives: minimizing the number of genes and maximizing the classification accuracy. MOBHHO uses two fitness functions to solve competing objectives. The first function based on SVM with LOOCV classifier and the second function also depends on KNN with K-fold classifier, as well as the percentage of gene selection found in both functions. Furthermore, MOBHHO tries to find the Pareto-optimal solutions, i.e. the best gene subset that contains a minimal number of selected genes and better classification accuracy. We have integrated several filter-based ranking methods with our proposal. In order to test the performance accuracy of the proposed MOBHHO algorithm, we compared our algorithm with other recently published algorithms in the literature. The experiment results which have been conducted on eight benchmarks (binary-class and multi-class), MOBHHO able to provide a minimum number of genes to obtain the highest classification accuracy. The proposed method reaches above 98% classification accuracy in six benchmark datasets and a maximum accuracy of 100% is achieved.