K-means聚类分析在人体体型分类中的应用

Application of K-means Clustering Analysis in the Body Shape Classification

讨论了K－means聚类分析在人体体型分类应用时分类数的确定方法和迭代收敛两个重要问题．参考GB／T1335—2008，以219名青年女性人体数据为检验样本，以胸腰差为实例进行论证．结果表明：采用基于系统聚类的距离评价函数法，样本最佳分类数为7类，如限制分类数为3～5时，则最优分类数为4；抽取容量分别为219和10O的两组样本进行不同迭代次数的聚类分析，发现聚类收敛所需的迭代次数受数据离散程度影响，采用SPSS软件进行聚类分析时应该设定较大的迭代次数以确保聚类收敛．

The determination of the classification number and the convergence of the iteration were discussed, which were important issues of K-means clustering in the classification of human body shape. GB/T 1335—2008 was used as reference standard, 219 young female body data were used as test samples, bust-waist was the variable for demonstration. The distance evaluation function based on hierarchical clustering method was put forward to explore the clustering number of the body data, it was showed that the optimal clustering number is 7, when the clustering number limited 3-5 classes, 4 is optimal. Different iteration numbers were analyzed in the process of clustering, using two different sets of data, one is 219 samples the other is 100 samples. The result showed that iterations numbers required for convergence was affected by the degree of discrete data, so iteration number should be bigger in SPSS to ensure convergence.