Optimizing Data Classification in Support Vector Machines Using Metaheuristic Algorithms

Abstract

To categorize patient diagnosis data related to Chronic Kidney Disease (CKD), this study compares the classification performance of Support Vector Machines (SVM) enhanced by Particle Swarm Optimization (PSO) and Genetic Algorithm (GA). CKD is a severe illness in which the kidneys fail to adequately filter blood and perform their normal functions. This study utilized secondary data consisting of patient conditions and health information. Based on references from CKD-related journals, 15 independent variables and one dependent variable were selected from an initial set of 54 variables. To address the issue of unbalanced data, an oversampling technique was applied, and the data was subsequently split into 80% for training and 20% for testing. During the training phase, SVM-PSO and SVM-GA models were developed, and the gamma value was optimized using the RBF kernel function of SVM. The results indicated that in classifying CKD patient diagnosis data, the SVM-PSO model (97.54% accuracy) outperformed the SVM-GA model (97.37% accuracy). This finding suggests that PSO-based hyperparameter optimization yields a superior model for data classification