KDRuleEx: A Novel Approach for Enhancing User Comprehensibility Using Rule Extraction

Knowledge representation opaque model like ANN has advantage of accuracy and limitation of interpretability. Transparent models like decision tree, decision table and rules represent knowledge in more understandable form and can easily be integrated with other learning system. Converting an opaque model like ANN to transparent model is called rule extraction. Lot of work has been done in the field of rule extraction like "SVM to Rules", "ANN to Decision Tree", "ANN to Rules" etc. We did not find any direct approach of converting ANN to decision table in our literature survey. In this paper, we proposed a novel pedagogical rule extraction technique to generate a decision table using training example set and a trained artificial neural network on it. The proposed algorithm can be used with both discrete as well as continuous input. The proposed algorithm has advantage in terms of computational performance and memory management.