一种融合了异常数据识别的CMM改进算法

摘要

针对聚类过程中有意义的异常数据难以识别的问题，在改进CMM算法的基础上，提出了一种融合了异常数据识别的层次聚类算法。采用CMM方法提出的原子簇思想，通过重新定义簇中心、噪声判断标准以及改进循环机制等手段提高聚类准确性及算法效率。提出了异常数据的概念和定义，并将其识别算法引入聚类过程过程。基于仿真及实际数据的实验结果证明，该算法能够根据设定参数准确识别异常数据，同时其聚类准确性及性能针对CMM算法也有了相应提高。%A hierarchical clustering algorithm with abnormal data recognition, CDCMM algorithm, is introduced regarding the problem of recognizing meaningful data which is non-mainstream in mining process. Based on the concept of atomic clustering from CMM algorithm, CDCMM improves clustering accuracy and performance by redefining the concept of cluster medoid and the standard used to distinguish noise data, and improving the circulation mechanism. The concept and definition of abnormal data are introduced and its recognition algorithm is added to clustering procedure. Experimental results on both simulation and produc-tion dataset show that the proposed algorithm can recognize the abnormal correctly data with responding parameters, and obtain higher clustering accuracy and performance.