The Fuzzy Decision Tree Algorithm Based On Dynamic Partition Of The Feature Space

Decision Tree classification algorithm is one of the most widely used algorithms in the field of pattern recognition, by virtue of its high classification accuracy and strong interpretability and understandability of the classification results. At present, the decision tree classification technology has been successfully applied to the business decision analysis, medical diagnosis, robot technology and many other fields.Firstly, this paper introduces the technology of data mining and pattern recognition, including Bayes classification algorithm, support vector machine(SVM) algorithm, K nearest neighbor algorithm, artificial network algorithm and decision tree algorithm. By analyzing different decision tree algorithms and summarizing the main procedure on the basis of comparing the different decision tree algorithms, this paper proposes a new fuzzy decision tree algorithm, which dynamically partitions the feature space during the period of constructing tree, instead of partitioning ahead of the process of constructing tree like other algorithms. When making fuzzy partition for the current nodes, this algorithm uses fuzzy c-means to cluster the sample, then assesses each attribute in terms of the information gain, and finally chooses the attribute corresponding to the biggest information gain to make fuzzy partition with the current node. In addition, the constructing tree pattern in the new algorithm highlights the optimal partition in every node, which will surely result in the problem of over-training. Pro-pruning strategy is employed to solve that problem, which can not only improve the test accuracy of the decision tree but also control the magnitude of decision tree and improve the interpretability.Furthermore,15data of UCI data library are applied to test the invalidity of the proposed algorithm by comparing the results with fuzzy FDTs algorithm, C4.5, CART and FS-DT fuzzy decision tree. The test results that the proposed fuzzy decision tree algorithm is better than the most of the mentioned classifiers in terms of accuracy and simple model.