Title :
Hybrid rule-extraction from support vector machines
Author :
Diederich, J. ; Barakat, N.
Author_Institution :
Fac. of Appl. Sci., Sohar Univ.
Abstract :
Rule-extraction from artificial neural networks (ANNs) as well as support vector machines (SVMs) provide explanations for the decisions made by these systems. This explanation capability is very important in applications such as medical diagnosis. Over the last decade, a multitude of algorithms for rule-extraction from ANNs have been developed. However, rule-extraction from SVMs is not widely available yet. In this paper, a hybrid approach for rule-extraction from SVMs is outlined. This approach has two basic components: (1) data reduction using a logistic regression model and (2) learning based rule-extraction. The quality of the extracted rules is then evaluated in terms of fidelity, accuracy, consistency and comprehensibility. The rules are also verified against the available knowledge from the domain problem (diabetes) to assure correctness and validity
Keywords :
data reduction; knowledge acquisition; regression analysis; support vector machines; SVM; artificial neural network; data mining; data reduction; hybrid computational intelligence algorithm; learning based rule-extraction; logistic regression; support vector machine; Artificial intelligence; Artificial neural networks; Australia; Data mining; Diabetes; Information technology; Machine learning algorithms; Medical diagnosis; Support vector machine classification; Support vector machines;
Conference_Titel :
Cybernetics and Intelligent Systems, 2004 IEEE Conference on
Conference_Location :
Singapore
Print_ISBN :
0-7803-8643-4
DOI :
10.1109/ICCIS.2004.1460774
