改进的混合属性数据聚类算法

Improved clustering algorithm for mixture data sets

k-prototypes是目前处理数值属性和分类属性混合数据主要的聚类算法,但其聚类结果对初值有明显的依赖性。对k-prototypes初值选取方法进行了分析和研究,提出一种新的改进方法。该方法有更高的稳定性和较强的伸缩性,可减少一定程度的上随机性。实际数据集仿真结果表明,改进算法是正确和有效的。

The k-prototypes algorithm has become popular technique in solving mixed numeric and categorical data clustering problems in different application domains. However, it requires random selection of initial points for the clusters. So it is obvious that outputs are especially sensitive to initial. Different initial points often lead to considerable distinct clustering results. The method of random selection is analysed and a method of searching initial starting points is proposed through grouping data sets. Experiments show that new initialization method leads to better accurate and scalable.