The identification of fuzzy weighted classification system incorporated with Fuzzy Naive Bayes from data

Abstract

We present an effective approach for identifying the rule base, the weight of each proposition and the certainty factor of each rule in the fuzzy weighted classification system. Here we use the idea of Fuzzy Naive Bayes, and assume variables in antecedents are conditional independent given corresponding consequents. So the weight of each proposition in the antecedent can be computed out via the conditional probability of the corresponding consequent given this proposition in the antecedent, and the certainty factor of each rule can also be computed out via the conditional probability of the consequent in this rule given the corresponding antecedent. The final rule base consists of all fuzzy rules whose certainty factors are large enough. Using this approach we can get a concise and consistent fuzzy system. Before computing all kinds of weights according to the proposed method, an optimal fuzzy partition is achieved by using fuzzy clustering, for instance, fuzzy c-means or similar cluster method. Finally we evaluate our identification method using Fisher iris data and Diabetes data, the tests show that a high recognition rate can be achieved.