A NEW CLUSTERING-BASED APPROACH FOR MODELING FUZZYRULE-BASED CLASSIFICATION SYSTEMS

In the present study, we propose a novel clustering-based method for modeling accuratefuzzy rule-based classification systems. The new method is a combination of a data mappingmethod, subtractive clustering method and an efficient gradient descent algorithm. A data mappingmethod considers the intricate geometric relationships that may exist among the data and computesa new representation of data that optimally preserves local neighbourhood information in a certainsense. The approach uses subtractive clustering method to extract the fuzzy classification rulesfrom data; the rule parameters are then optimized by using an efficient gradient descent algorithm.Twenty datasets taken from UCI repository are employed to compare the performance of theproposed approach with the other similar existing classifiers. Some non-parametric statistical testsare utilized to compare the results obtained in experiments. The statistical comparisons confirm thesuperiority of the proposed method compared to other similar classifiers, both in terms ofclassification accuracy and computational effort.