Extended Taxonomy of Rule Extraction Techniques and Assessment of KDRuleEx

Classifiers like ANN & SVM are always preferred over other classification model like decision tree due to higher accuracy but lacking explainability and comprehensibility. Rule extraction techniques bridges gap between accuracy and comprehensibility. To evaluate and compare different rule extraction techniques, we require measures for evaluation and categorization. Taxonomy helps us to select a technique based on the requirements and desired priorities. In this paper, we extended popular ADT-taxonomy of rule extraction which has been designed for ANN as underlying model. Proposed taxonomy covers all types of work related to rule extraction. It makes easier to introduce new rule extraction techniques by improving the performance on evaluation criteria. In this paper almost all possible aspects of evaluation and categorization of rule extraction techniques has been considered and further used to evaluate the algorithm KDRuleEx.