A Comparative Study On CRI And Similarity-Based Reasoning For Classification

Since Zadeh proposed the fuzzy sets theory in 1965, fuzzy reasoning as the foundation of the application of fuzzy sets theory, is the scholars'study focus all the times. Fuzzy production rule is the mostly used knowledge representation form in the fuzzy reasoning as its explicit expression, strong readability and interpretable ability. There are mainly two methods in the production rules-based fuzzy reasoning. One method proposed by Zadeh is CRI method. The other is the similarity-based reasoning method. The paper mainly discusses the application of CRI method, and the comparison of CRI method and similarity-based method in the fuzzy classification problems.The paper firstly introduces and analyses CRI method and similarity-based method in detail. Then it discusses the application of CRI method in the fuzzy classification problems in the following three aspects: (1) the influence of different implication operators on the classification result; (2) the influence of different shape of triangles on the classification result when fuzzying the example as a triangle; (3) when matching the corresponding attributes of an example to the attributes of the antecedent part in a rule respectively, the paper uses the max(min) and the maximum information entropy methods to calculate the classification result for this rule. And the paper compares them in the reasoning mechanism and reasoning process in the fuzzy classification problems.Finally the paper tests the classification accuracy of the two methods in the Pima, Iris, Glass, Wine databases. The experiments show that choosing proper parameters can make the classification accuracy obtained by CRI method better than the accuracy obtained by similarity-based method. Thereby proves the validity of CRI method in the classification problems.