红斑鳞状皮肤病的聚类分析

Clustering Analysis for Erythemato-Squamous Diseases

针对红斑鳞状皮肤病鉴别诊断难题,提出利用聚类分析进行诊断;采用3种k-均值、2种k-中心点、最小生成树以及密度峰值点快速搜索聚类算法对该疾病数据进行分析,比较各算法对该疾病的聚类误差平方和、聚类结果 Rand指数、Jaccard系数、调整Rand指数以及聚类准确率;比较各算法对预处理的数据子集的聚类结果与文献中采用k-均值算法对未预处理的该数据子集的聚类结果。结果表明:邻域k-中心点算法对红斑鳞状皮肤病有很好的聚类效果,聚类准确率、聚类结果 Rand指数、Jaccard系数、调整Rand指数均优于对比算法,密度全局k-均值算法的聚类效果次之,全局k-均值算法取得最佳聚类误差平方和;k-均值算法对预处理数据子集的聚类准确率最高,邻域k-中心点与密度全局k-均值算法的聚类准确率相等;数据预处理可提高k-均值算法对该疾病的诊断准确率。

To solve the diagnosis problem of erythemato-squamous diseases, the clustering analysis method was suggested to diagnose. Three k-means algorithms, two k-medoids algorithms, minimum spanning tree and the clustering by fast search and find of density peaks algorithms were applied to analyze the diseases data. The clustering results of these algorithms were compared in terms of the clustering error, Rand index, Jaccard coefficient, adjusted Rand index, and clustering accuracy. The clustering results of these algorithms on the pre-processed subset of the diseases dataset were compared with those of k-means algorithms on the original subset. The results show that the neighborhood based k-medoids algorithm gets the best clustering results in terms of the clustering accuracy, Rand index, Jaccard coefficient and adjusted Rand index. The density based global k-means algorithm follows. The global k-means algorithm gets the optimal clustering error. The clustering accuracy of k-means algorithms on the pre-processed subset is superior to others. The neighbor- hood based k-medoids and density global k-means algorithm get the same clustering accuracy, which is inferior to that in the literature. The preprocess of the dataset can improve the diagnosis accuracy of k-means algorithms to the diseases.