Comparison and Performance Analysis of SVM and PSO-SVM Algorithms (Case Study Classification of Senior High School)

Attribute Selection is very important for classification process. This research has been done by doing attribute selection using PSO method (Particle Swarm Optimization) on SVM algorithm (Support Vector Machine). The development of the classification model uses three parameters especially data attribute, influence of the transformation of various kernel function and penalty factor (C) toward the performance of SVM and PSO-SVM classification. The analysis uses five kernels in mySVM library that existed in Rapidminer application namely dot, radial, polynomial, neural, and anova kernel. The training data used in the first model classification development is student interest data at ABC high school on 2013-2014 year academic. The first model is evaluated using accuracy, precision, recall, and auc value test. The first result shows that the anova kernel on PSO- SVM is able to work with accuracy level 99.30% using penalty factor 0.1. The second model has been developed to predict student interest in XYZ high school. The second result shows that PSO-SVM with kernel anova is able to classify students interest with 99.29% accuracy level.