摘要
针对传统K-prototypes在计算分类属性的差异度时未考虑各个分类属性对聚类结果的影响程度,且算法容易受到噪声的干扰,无法处理数据中不够精确、不完整等不确定性问题,提出基于信息熵的粗糙K-prototypes聚类算法。在计算数据样本之间分类属性的差异度时,使用信息熵的理论,确定每个分类属性对于聚类分析结果的影响权重;引入粗糙理论,计算得到各样本与粗糙模之间的粗糙相异度,通过多次迭代计算,获得最终聚类结果。该算法结合信息熵和粗糙理论,可区别对待各分类属性,解决数据不精确引起的不确定性问题,4个UCI数据集上的实验分析结果验证了该算法的有效性。
The traditional K-prototypes fail to concern the degree that every category attribution effecting clustering results,and they are easily disturbed by noise and can not deal with these uncertain problems including inaccuracy and incompleteness.To solve these faults,an entropy based rough K-prototypes clustering algorithm was provided.Firstly,using entropy theory,the weight of every category attribution for clustering result was got,and then using rough theory,the rough dissimilarity degree be-tween every sample and every rough model was calculated.Through multi iterative calculations,the final clustering results were got.The new method combines information entropy theory and rough set theory,which can differentiate every category attribu-tion and cope with uncertain problems caused by inaccuracy data.Results of experiments on four UCI data sets show the algo-rithm is effective.
出处
《计算机工程与设计》
北大核心
2015年第5期1239-1243,共5页
Computer Engineering and Design
基金
国家自然科学基金项目(61364006)
广西自然科学基金项目(2013GXNSFAA019336
2013GXNSFBA019280)
广西高校科学技术研究基金项目(LX2014190)
广西科技大学科学基金项目(校科自1261128)
关键词
混合型数据
聚类
信息熵
粗糙集
数据挖掘
mixed data
clustering
information entropy
rough set
data mining
