A Clustering Analysis Method Based on Wilcoxon-Mann-Whitney Testing

As the core step of clustering analysis, the results of distance measurements will influence the clustering accuracy. The existing measurements are mostly based on the information about cluster features. However, the cluster features may be not sufficient enough and would result in losing data information about clusters containing a number of objects. To improve the measurement accuracy, we make full use of the distribution characteristics of objects in clusters, so we use the descriptive statistics and the Wilcoxon-Mann-Whitney rank sum test in nonparametric statistics to measure distances during clustering. Furthermore, a two-stage clustering is proposed to improve the performance of clustering analysis, from the aspects ofavoiding assuming the number of clusterspreliminarily, discovering clusters of arbitrary shapes andimproving clustering accuracy. The experiments on multiple datasets compared with other clustering algorithms illustrate the accuracy and efficiency of the proposed clustering algorithm.