Extracting Symbolic Rules from Trained Support Vector Machines Based on the Derivative Heuristic Information

Support vector machines (SVMs) have demonstrated superior performance compared to other machine learning techniques, especially in classification problems. How to extract rules from trained SVMs has become an important preprocessing technique for data mining, pattern classification, and so on. In this paper, a new approach for symbolic rule extraction from trained SVMs is proposed. It includes the methods of the attribute selection, the division of attribute space, the rule expression, and so on. A new algorithm for rule extraction is given. The performance of the new approach is demonstrated by several computing cases. Experiment results show that the proposed approach can improve the validity of the extracted rules remarkably compared to other rule extracting approaches, especially for complicated classification problems.