Research And Improvement Of Cluster-Oriented Fuzzy Decision Trees

Decision tree learning is one of the most commonly used methods of machine learning. It attempts to learn a mathematical model to describe a sample set, in which the samples are disorderly and unsystematic. It is noted that, traditional decision tree algorithms (such as ID3) are often used to handle samples with discreteattributes. When there are numerical attributes, the step of discretizing attributes is a must. However, the discretization mechanism may lead to information loss, and then affects the performance of the generated decision tree. In recent years, the algorithms of building a cluster-oriented fuzzy decision trees have been presented in order to avoid the discretization of numerical attributes. This type of algorithms combines fuzzy clustering and decision tree learning into together, and the decision trees grow based on fuzzy clusters.In this thesis, we firstly study the architecture and growth mechanism of the cluster-oriented fuzzy decision tree, and analyze the termination criterion of node expanding. Secondly, since the traditional cluster-oriented fuzzy decision trees could not deal with the unlabelled samples, the thesis presents a new unsupervised cluster-oriented fuzzy decision tree algorithm. Based on the unlabelled samples, a new splitting criterion is proposed to expand each node, and then an unsupervised cluster-oriented decision tree is generated. Thus the method of cluster-oriented fuzzy decision tree can be generalized to the unsupervised learning domain. Finally, some experiments are conducted on synthetic and machine learning data sets. The experimental results show that the proposed method can achieve better performance comparing to the standard decision tree (C4.5) and traditional cluster-oriented fuzzy decision tree, and the size of its generated decision tree is the smallest one.