查看文章

Nowadays, with the increasing availability of online text documents, it becomes an important task for an organization to automatically classify the document. In Text Classification (TC), Support Vector Machine is the commonly used machine-learning algorithm. Performance of SVM highly depends on parameter tuning using metaheuristic algorithm for text classification. To integrate dynamic searching to parameter setting for SVM is a big issue that produced great influence in the classification accuracy. In order to improve the generalization and learning capability of SVM, this paper presents a new approach known as RSS-SVM, which is used to optimize kernel function and penalty parameters through the Ringed Seal Search algorithm. Experiments are conducted on three text datasets named: Reuter21578, 20 newsgroup and TDT2 with a different number of classes, which shows that proposed RSS-SVM present significant results having 79.22% accuracy, 70.79% recall, 58% precision and 54.71% fmeasure among the previous GA-SVM and CS-SVM algorithms.