摘要
在k-means算法基础上,提出利用平均相对偏差对数据的维分布密集度进行度量,并根据空间分布的密集度动态地给属性赋予权值。在计算平均相对偏差时,度量值与平均值间的偏差没有被平方,在一定程度上降低了孤立点的影响,与标准差相比具有更强的鲁棒性。仿真结果表明,基于平均相对偏差的聚类算法提高了聚类的质量。
Based on the k-means algorithm, mean relative deviation is presented to measure the degree of denseness, and dynamic weights of variables are associated with it. Because deviation between data and average is not squared, the effect of isolated data is reduced, which is more robustness than standard deviation. The simulation results show the mean relative deviation based on clustering algorithm could improve the quality of clustering.
出处
《兵工自动化》
2008年第8期32-34,共3页
Ordnance Industry Automation