多维空间可调整的近邻传播聚类算法

Multidimensional Spatial Adjustable Affinity Propagation Clustering Algorithm

针对近邻传播算法不适合处理多重尺度和任意形状数据的问题,提出了一种基于多维空间可变换的MSAAP(multidimensional similarity adaptive affinity propagation)算法。首先,通过熵值法计算数据样本点的属性权重;然后,根据属性权重构造出一种新型计算相似性矩阵的方法;最后,根据属性权重的优先级将样本点的空间划分成若干个空间块,并计算空间块的吸引度和归属度之和,进而调整样本点的空间分布。通过13个不同形状的UCI数据集和3个人脸数据库进行对比实验,从准确率、算法时间、聚类个数3个维度去分析,最终实验结果证明所提出的MSAAP算法聚类效果更优。

In view of the problem that the affinity propagation algorithm is not suitable for dealing with multiple scales and arbitrary shape data, this paper presents an MSAAP(multidimensional similarity adaptive affinity propagation) algorithm based on multidimensional spatial transform. First, the attribute weight of the data set is calculated by the entropy method. Second, this paper puts forward a new method of computing the similarity matrix based on attribute weights. Last, this paper divides the spatial space of the sample points into several spatial blocks according to the priority of the attribute weights, calculates the degree of appeal and the degree of attribution of the space block, and then adjusts the spatial distribution of particles. This paper uses 13 UCI data sets and 3 face databases to do comparative experiments, and then from the accuracy rate, the algorithm time, the number of clusters to analyze the 3 dimensions. Finally, the experimental results show that the MSAAP clustering effect is better.