Bio-Inspired Feature Selection Algorithms With Their Applications: A Systematic Literature Review

Based on the principles of the biological evolution of nature, bio-inspired algorithms are gaining popularity in developing robust techniques for optimization. Unlike gradient descent optimization methods, these metaheuristic algorithms are computationally less expensive, and can also considerably perform well with nonlinear and high-dimensional data. Objectives: To understand the algorithms, application domains, effectiveness, and challenges of bio-inspired feature selection techniques. Method: A systematic literature review is conducted on five major digital databases of science and engineering. Results: The primary search included 695 articles. After removing 263 duplicated articles, 432 studies remained to be screened. Among those, 317 irrelevant papers were removed. We then excluded 77 studies according to the exclusion criteria. Finally, 38 articles were selected for this study. Conclusion: Out of 38 studies, 28 papers discussed Swarm-based algorithms, 2 papers studied Genetic Algorithms, and 8 papers covered algorithms in both categories. Considering the application domains, 21 of the articles focused on problems in the healthcare sector, while the rest mainly investigated issues in cybersecurity, text classification, and image processing. Hybridization with other BIAs was employed by approximately 18.5% of papers, and 13 out of 38 studies used S-shaped transfer functions. The majority of studies used supervised classification methods such as k-NN and SVM for building fitness functions. Accordingly, we conclude that future research should focus on applying bio-inspired feature selection to a diverse area of applications such as finance and social networks. And further exploration into enhancement techniques such as quantum representation, rough set theory, chaotic maps, and Lévy flight is necessary. Additionally, we suggest investigating other transfer functions besides S-shaped, such as V-shaped and X-shaped. Moreover, clustering and deep learning models for constructing fitness functions in bio-inspired feature selection algorithms need to be investigated further.