JoWUA

Abstract

Data analysis involves feature selection to achieve better performance, increase the speed of the learning algorithm, and get more precise and comprehensible results. Physics models have been widely used to accelerate the search process over the solution space. Moreover, the flexibility of the search process enables flexibility in adapting to the specific problem at hand, for example, by tuning parameters to optimize feature selection for various kinds of problems, or datasets, or learning tasks. The paper investigates the Atomic Orbital Search (AOS) algorithm, which is based on the quantum mechanical formation of electron orbitals, for feature selection. According to the PRISMA 2020 guidelines, literature published from 2019 to 25 October was collected from the Scopus and Web of Science databases. After duplicate removal and multi-stage screening, only 34 studies that strictly focused on AOS algorithms and met the inclusion criteria were included in the qualitative and quantitative synthesis. This paper is a summary of the present developments and applications of the Atomic Orbital Search (AOS) algorithm. The literature review indicates that the publications for the AOS algorithm have grown from 1 in 2019 to 11 in 2024, with the highest number of citations in 2021, 336. Empirical studies reveal that the accuracy of the AOS-based models is always higher than that of any other models, and feature reduction is always greater than 50% with the classification accuracy always higher than reported by several studies. The AOS-based models also have certain drawbacks, such as parameter sensitivity, simulation-verification, and the lack of comparison with other models, including performance indicators and applicability to various datasets, which can make them less effective in practical use and reduce their applicability to a variety of situations. This review focuses on the feature selection using AOS and other physics-inspired optimization methods.