The Study On The Heuristic Algorithm For Generation Of Fuzzy Decision Tree

Fuzzy decision tree learning is a very efficient and practical method on account of merging the advantages of fuzzy sets theory and decision tree learning. It has been successfully applied to a broad range of knowledge acquisitions from diagnose medical to assess to credit risk of loan applicant.However, constructing the optimal fuzzy decision tree has been proved to be NP-hard, a heuristic algorithm is necessary. FUZZY -ID3 algorithm is the most popular and the most efficient heuristic algorithm in the existing references, the heuristic information in which is fuzzy information gain, i.e., the fuzzy mutual information between the attributes and the class. It makes the fuzzy information entropy converge quickly, and we can obtain a small tree without much computation. But almost all heuristic algorithms ignore the relevance (redundancy) between attributes. Therefore, after analyzing the effect of relevance between attributes on the generation of fuzzy decision tree and the shortcoming of FUZZY - IDS in detail, we propose an extended algorithm to FUZZY -ID3. An updated heuristic is adapted, which aims at selecting the attribute that can bring not only the more mutual information between a candidate attribute and the class, but also the less mutual information between a candidate attribute and the selected attributes on the same branch as the next testing attribute. The extension is based on adding the process of reducing the effect of redundancy between attributes to generation of fuzzy decision tree.Numerous figures and tables from the experiment data are cited to illustrate the difference between the extended heuristic algorithm and FUZZY-ID3. The result turns out to be two situations: one is that, to some datasets, the extended algorithm is superior to Fuzzy-ID3, and in another situation, the two are equal.