基于凝聚函数的混合属性数据聚类算法

借助于近似极大值函数的凝聚函数，将传统数据聚类问题转化为无约束优化问题求解．首先利用一阶必要条件，推导出数值属性下数据聚共中心的计算格式;其次采用类属性分解方法，提出计算类属性数据对象之间距离的新方法，井在此基础上给出混合属性下数据聚类中心的计算格式和一个能处理数值型和分类型混合数据集的凝聚聚类算法；最后选取不同初始聚类中心，使用凝聚聚类算法对英语借词进行了聚类实验和分析．结果表明，凝聚聚类算法在计算效率和计算效果方面均优于模糊k-prototypes聚类算法．

A clustering algorithm for mixed valued data based on aggregate function

Aggregate function which approximates the maximum function, is introduced, and data clustering problem is reformulated as the unconstrained optimization. Firstly, a computing scheme for clustering center is inferred for numeric valued data, applying the first order necessary condition. Secondly, a new distance concept and computing scheme for categorical valued data are presented using decomposition method of categorical valued attributes, and furthermore, a new clustering approach for mixed numeric and categorical valued data is presented. Finally, computing experiment and analysis for Chinese loanwords in English are given by using different centers of clustering. The results show that the aggregate clustering algorithm is superior to the fuzzy k-prototypes algorithm in both computing efficiency and effects.