Optimizing Machine Learning Models Performance Using Swarm-Based Feature Selection: Methods and Experimental Evaluation

Resumen

Abstract: Large-scale datasets have grown rapidly in recent decades due to the rapid advancement of database and computer technology. However, the number of data mining applications that use high-dimensional datasets and need great speed and accuracy is growing quickly. The curse of dimensionality, in which there are substantially more characteristics than patterns, is a key problem with these applications. Feature selection is one of the dimensionality reduction strategies that can improve the data mining task's accuracy and lower its computational cost. The goal of the feature selection process is to choose a subset of features that are most relevant to the target class and have the lowest inner similarity. By removing unnecessary, redundant, or noisy data, it lowers the dimensionality of the data. This study compares and contrasts several feature selection techniques and provides a general classification of them. This research also examines the latest swarm intelligence techniques and reviews feature selection techniques based on these algorithms using the credit card fraud detection dataset. Additionally, we assess the advantages and disadvantages of the various swarm intelligence-based feature selection techniques that were examined.

Keywords: Swarm intelligence, evolutionary algorithms, feature selection, machine learning, and dimensionality reduction.