Rule extraction using Support Vector Machine based hybrid classifier
M.A.H. Farquad, Vadlamani Ravi, Raju Surampudi Bapi · 2008
Support Vector Machines (SVMs) have become an increasingly popular tool for machine learning tasks involving classification and regression, and have shown superior performance compared to other machine learning techniques. In this paper we propose a hybrid classification technique to extract fuzzy rules from the support vector machine and evaluate the rules against decision tree classifier constructed from the same support vector machine. The hybrid approach proceeds in three major steps. In the first step we use training patterns with class labels to build an SVM model that gives the support vectors with acceptable accuracy. Fuzzy rules are generated using the extracted support vectors during second step. In the final step the resulting rule set is tested using the test data of the problem. The quality of the extracted rules is then evaluated in terms of accuracy and fidelity. It is found that the proposed hybrid approach using fuzzy rules yielded highest accuracy and fidelity compared to hybrid with decision tree classifier.