Enhance the Performance of Independent Component Analysis for Text Classification by Using Particle Swarm Optimization

Abstract

Independent component analysis is a statistical model that is used to separate a multivariate signal into additive components. Independent component analysis has gained much attention in recent years in the neural networks and signals processing fields. Several data mining applications with Independent component analysis have been considered, such as latent variable decomposition, analysis of text document data, detection of hidden signals in satellite imagery, and weather data mining. The conventional Independent component analysis search scheme is based on a gradient algorithm, which requires a predefined learning rate. Therefore, it cannot solve the convergence dilemma. To overwhelm the disadvantage, particle swarm optimization is employed in the ICA algorithm. In statistics, negentropy is used as a measure of distance to normality. The present study used a metaheuristic, particle swarm optimization algorithm that employs negentropy as a fitness function to enhance the performance of independent component analysis for the text classification model as one of the text mining applications. The proposed system was applied to a medical corpus, and two experiments were executed. Results show that the performance of the PSO-ICA algorithm is superior to the FastICA for text classification, where it achieves an overall F -measure of 89% for text classification compared with the FastICA algorithm, which provides 85% of an overall F -measure for text classification.