ENHANCING CLASSIFICATION PERFORMANCE OF SVM USING HYBRID PSO ALGORITHM: A CASE STUDY ON LEARNING DISABILITY DIAGNOSIS

Abstract

Learning disabilities are conditions that impact an individual's capacity to comprehend or employ spoken or written languages. This research paper, it is aimed to classify students with and without learning disabilities using machine learning techniques. A real dataset of 348 students was used for this purpose. It was found that traditional classification methods such as decision tree, k-nearest neighbors, etc. achieved an accuracy of over 90%, while the support vector machine (SVM) performed poorly with only 80% accuracy.

In this paper, the main objective is to improve SVM by using the advantage of optimization techniques. Firstly, a grid search algorithm was applied, resulting in an accuracy of 92.7%. Furthermore, a global best PSO algorithm was employed, which improved the accuracy to 94.5%, and a local best PSO algorithm, which improved the accuracy to 95.4%. Finally, a new hybrid PSO algorithm that combined the advantages of both local and global PSO was applied, resulting in an accuracy of 97.27%.

The outcomes indicate that the hybrid PSO-SVM algorithm is a highly efficient approach for distinguishing between students with learning disabilities and those without. This could be of great help to educators in identifying students who may require additional support in their studies. Additionally, the study highlights the limitations of traditional classification methods and the potential of machine learning techniques, specifically the use of optimization algorithms with SVM, in improving classification accuracy.