The possibilistic tree classifier, fuzzy classification trees, has been introduced to discover the knowledge in large amount of data. Traditional tree classifiers often make a single decision for classifications. A single decision is too overly general for knowledge discovery, especially to some application domains, such as medicine and finance. Fuzzy classifications have been proposed to solve such a serious problem. Instead of rigidly determining a single class for any given instance, fuzzy classifica-tion trees give predictions about the degree of possibility for every class. In order to select the best attribute for each step of the induction task, possibilistic entropy evaluation addressed in this paper is to evaluate the uncertainties in the data. By comparing with decision tree classifier, this paper also shows that the classifier generated in accordance with the concept of fuzzy classifications is a more gen-eral model than the traditional tree classifier. The fuzzy classification trees have been applied to some data sets from the UCI repository. Generally speaking from empirical results, the misclassification rates of those data sets are less than C4.5.
展开▼
