A Comparison Study on Particle Swarm Optimization (PSO) Algorithms for Data Feature Selection

High-dimensional data presents a challenge for big data analysis and machine learning due to its complexity and size. To alleviate the pain of handling such data, feature selection is a good choice to choose the most relevant and useful features. To do so, there are several strategies to choose from and this work focuses on the Particle Swarm Optimization (PSO) for classifying to famous datasets: the Arrhythmia and the DBWorld e-mails datasets. For the sake of comparison, five PSO algorithms containing different weights are paid attention to and three well-known classifiers namely: decision tree (DT), random forest (RF), and support vector machine (SVM) are performed. The results show that all five PSOs perform well using only approximately 50% of the dataset. Evidently, a “weight” parameter plays a crucial role in the PSO algorithm and clearly deserves further investigation. With a good combination between the weight and the classifier, our experiments indicate that satisfactory results can be obtained.