基于遗传算法的混合属性聚类初始点选择研究  

Initial Points Chosen in Mixture Data Clustering Algorithm Improvement Based on Genetic

作　　者：赵立江[1] 

机构地区：[1]徐州师范大学计算机学院,江苏徐州221116

出　　处：《广西师范大学学报（自然科学版）》2008年第3期194-197,共4页Journal of Guangxi Normal University:Natural Science Edition

基　　金：国家自然科学基金资助项目(70171033);江苏省高校自然科学基础研究资助项目(07KJDS520216)

摘　　要：k-prototypes是处理包含数值属性和分类属性混合数据主要的聚类算法,但由于对初始点的随机选择极易使聚类结果陷入局部最优,导致聚类结果的偏差。为改善聚类结果对初始点的依赖,采用遗传算法对初始点进行全局最优搜索。改进后的方法有更高的稳定性和对大数据集更快的收敛速度。数据集仿真结果表明改进算法正确、有效。The k-prototypes algorithm has become a popular technique in solving clustering problems for mixed numeric and categorical data in different application domains. However,random selection of initial points for the clusters was required. So it is obvious that outputs are especially sensitive to initial. Genetic algorithm is a method of searching for global optimal result. This paper analyses the method of random selection and proposes a method of searching initial starting points through genetic algorithm. Experiments show that the new initialization method proves to be more accurate and effective.

关 键 词：遗传算法 聚类 K-MEANS k-prototypes 分类型数据 相异度 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]