关键词: 归纳学习, 模糊决策树, 数据预处理, 模糊化, 粒子群优化算法

Abstract: Fuzzy Decision Tree induction（FDT） has been used in more and more application area. When the data are numerical, the FDT algorithms need to fuzzify them into some linguistic items. That how many linguistic items of an attribute are proper is not known. The selection generally depends on experts’ opinion or people’s common. Currently, it is not yet available to learn the number of linguistic items by using the experimental method of particle swarm optimization. The paper introduces a PSO based approach to optimize the selection of linguistic item’s number in fuzzifying processing of data in FDT（FDT-K algorithm）. Experimental studies show that the FDT-K algorithm compared with the people’s common methods to decide the number of linguistic items of attribute can create a better fuzzy decision tree with higher classification and generalization capability.

Key words: inductive learning, fuzzy decision tree, data preprocessing, fuzzification, particle swarm optimization