Knowledge Extraction from Piecewise-Linear Approximation of Multi-Surface Classifier
Международный молодежный форум «Радиоэлектроника и молодежь в XXI веке»
2012
Andrejs Bondarenko,
Arkādijs Borisovs
Current work is describing approach for interpretable if-then rules extraction from piecewise linear classifier. Such classifier can be built based on many black-box classifiers like RBF Neural Network, Multi Surface Method tree or can be created from scratch [1, 2]. Approach used to extract knowledge in the form of if-then rules from piecewise-linear classifier is extension of method proposed by [3] used to extract knowledge from linear SVM. We describe optimization problem, recursive algorithm used to extract if-then rules and its complexity. As well we present some experimental results highlighting proposed approach performance.
Atslēgas vārdi
Piece-wise linear approximation, knowledge extraction, polytope classifier
Bondarenko, A., Borisovs, A. Knowledge Extraction from Piecewise-Linear Approximation of Multi-Surface Classifier. No: Международный молодежный форум «Радиоэлектроника и молодежь в XXI веке», Ukraina, Харьков, 30.-30. janvāris, 2012. Харьков: Харьковский национальный университет радиоэлектроники , 2012, 9990.-9992.lpp.
Publikācijas valoda
English (en)