Piece-Wise Classifier Application to RBF Neural Network Rules Extraction

17th International Conference on Soft Computing (MENDEL'11) 2011
Andrejs Bondarenko, Tatjana Zmanovska, Arkādijs Borisovs

This paper describes the application of the symbolic knowledge extraction approach to RBF Neural Network (RBFNN). The proposed method retrieves a set of concise and interpretible IF-THEN rules from a piece-wise polytope classifier built on top of RBFNN. The majority of RBFNN rules extraction algorithms are dealing with neuron centers; we propose the same approach, but with the help of RBFNN transformation into a polytope classifier and application of recursive rule extraction algorithm. We present classification results of RBFNN extracted rules and show that under some cirumstances the polytope classifier extracted from RBFNN and rules extracted from this classifier are showing performance similar to RBFNN.

Atslēgas vārdi
Support Vector Machine, RBF, Rule Extraction, K-Means. Neural Network

Bondarenko, A., Zmanovska, T., Borisovs, A. Piece-Wise Classifier Application to RBF Neural Network Rules Extraction. No: 17th International Conference on Soft Computing (MENDEL'11), Čehija, Brno, 7.-7. maijs, 2011. Brno: Brno University of Technology, 2011, 9990.-9996.lpp.

Publikācijas valoda
English (en)