Towards an Improved Particle Swarm Optimization for Feature Selection: A Survey

Over the years, scientists have used natural discoveries such as evolution to solve real-world problems. Addressing the challenges that arise when dealing with high-dimensional data is one such problem. These challenges include difficulties in analyzing, visualizing, and modelling these high-dimensional data. As a result, the Swarm Intelligence (SI) techniquewas developed, which was inspired by natural swarm foraging behaviors. Particle swarm optimization (PSO) is a well-known SI algorithm for addressing a wide range of optimization problems. As a result, it has been used to solve a variety of optimization problems in fields as diverse as genomic analysis and intrusion detection systems. One of the most successful areas of PSO application is feature selection, which entails using computational techniques to select a reduced subset of features that have a sufficient relationship with their corresponding class labels. This, in turn, addresses the mentioned challenges. Nonetheless, progressive research has revealed several problems with PSO, including problems with diversity, and premature convergence among others. As a result, several improvements and extensions were made to various aspects of the algorithm since its inception to make it efficient. This paper organizes and summarizes current research on improvements to the PSO algorithm for solving the feature selection problem. Consequently, it presents current trends and directions for scholars in the field, as well as open challenges and literature gaps to investigate