混合属性数据聚类初始点选择的改进

Improved Clustering Algorithm for Mixed Numeric and Categorical Values

k-prototypes和模糊k-prototypes是处理数值属性和分类属性混合数据主要的聚类算法。但这两种聚类算法不足之处是对初值有明显的依赖。对初值选取方法进行了分析和研究,提出一种新的改进方法,可在一定程度上减少随机性。实际数据集仿真结果表明改进算法有更高的稳定性和较强的伸缩性。

The k-prototypes algorithm and Fuzzy k-prototypes algorithm have become popular technique in solving categorical data clustering problems in different application domains. However, they also reuires 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. This paper analyses the method of random selection and proposes a method of searching initial starting points through grouping data sets. Experiments show that the new initialization method leads to higher stability and flexibility.