Rule Extraction from DEWNN to Solve Classification and Regression Problems

Abstract

AbstractThis paper proposes a method to extract rules from differential evolution trained wavelet neural network (DEWNN) [1]. for solving classification and regression problems. The rule generation methods viz., Decision Tree (DT), Ripper and Classification and Regression Tree (CART) and Dynamic Evolving Neuro Fuzzy Inference System (DENFIS) are employed to extract rules from DEWNN for classification and regression problems respectively. The feature selection algorithm adapted by Chauhan et al., [1] is used in the present study. The effectiveness of the proposed hybrid is evaluated on Iris, Wine and four bankruptcy prediction datasets namely Spanish banks, Turkish banks, US banks, UK banks and Auto MPG dataset, Body fat dataset, Boston Housing dataset, Forest Fires dataset, Pollution dataset, by using 10-fold cross validation. From the results, it is concluded that the proposed hybrid method performed well in terms of sensitivity in classification problems.KeywordsDifferential Evolution Trained Wavelet Neural NetworkClassificationRegressionRule ExtractionDecision TreeRipperClassification and Regression TreeDynamic Evolving Neuro Fuzzy Inference System