摘要
针对高维数据的聚类过程不够直观、聚类结果也不易解释的问题,本文提出了一种基于改进雷达图的交互式可视化聚类方法。首先对传统雷达图进行了改进,采用熵权法确定数据的主要特征和属性排列,在去掉非主要特征基础上采用以极径表示属性值,以属性权重确定极角的改进雷达图进行数据可视化来突出数据的主要特征;然后采用改进的k-means算法对平面上的点集进行聚类,该改进算法不需事先给定簇的个数,能够依据密度和距离对初始中心进行优化,且在聚类过程中可交互调整参数,并使用不同颜色来区分不同类别,方便观察聚类过程和结果;最后通过仿真实验表明改进的雷达图更能反应数据的分布情况,改进的聚类算法具有更高的效率和聚类准确度。
A clustering method based on interactive visualization of radar chart is proposed.First,entropy-weight is adopted to determine the main features and arrange them.Removing the non-main feature from the original data,every data is described in an improved radar chart.The polar radius stands by the attribute weights.The polar angle stands by entropy-weight.Then the points in the radar chart are clustered through applying an improved-means algorithm.The number of clusters is not given before and the initial centers are optimized according to the point density and distance.Finally,the experiments show that the improved radar chart reflects the distribution of the data better and that the improved-means algorithm is more efficient and accuracy.
出处
《燕山大学学报》
CAS
2013年第1期58-62,共5页
Journal of Yanshan University
基金
国家自然科学基金资助项目(60970073)
