A Hybrid Approach Support Vector Machine (SVM) – Neuro Fuzzy For Fast Data Classification

Abstract

In recent decade, support vector machine (SVM) was a machine learning method that widely used in several application domains. It was due to SVM has a good performance for solving data classification problems, particularly in non-linear case. Nevertheless, several studies indicated that SVM still has some inadequacies, especially the high time complexity in testing phase that is caused by increasing the number of support vector for high dimensional data. To address this problem, we propose a hybrid approach SVM – Neuro Fuzzy (SVMNF), which neuro fuzzy here is used to avoid influence of support vector in testing phase of SVM. Moreover, our approach is also equipped with a feature selection that can reduce data attributes in testing phase, so that it can improve the effectiveness of time computation. Based on our evaluation in real benchmark datasets, our approach outperformed SVM in testing phase for solving data classification problems without significantly affecting the accuracy of SVM.