一种改进的GP-CLIQUE自适应高维子空间聚类算法

GP-CLIQUE算法是基于高斯过程的CLIQUE改进算法,但是此算法中存在固定划分网格和人为输入密度阈值这两个不足.提出了一种改进GP-CLIQUE的算法——REG-CLIQUE算法.该算法利用相对熵对每一维数据进行自适应网格划分,引入二叉树存储信息,删除冗余维,解决了固定划分网格的缺陷,提高了聚类结果的精度;同时,提出密度阈值的计算公式,根据数据集本身用递归算法计算密度阈值,解决了人为输入的不足,大大降低算法对先验知识的依赖性.实验结果表明,该算法在时间、聚类准确度等方面都优于GP-CLIQUE算法和CLIQUE算法.

An Improved GP-CLIQUE Adaptive High-Dimensional Subspace Clustering Algorithm

The GP-CLIQUE algorithm is an improved CLIQUE algorithm based on Gaussian process.However,there are two shortcomings of fixed partitioning grid and human input density threshold in this algorithm.An improved GP-CLIQUE algorithm,namely REG-CLIQUE was proposed.The relative entropy was used to carry on the adaptive meshing for each dimension data,the binary tree was introduced to store the information,the redundant dimension was deleted,the defect of fixed dividing grid was solved and the precision of the clustering result was improved.Meanwhile,the formula of density threshold was proposed,according to the data set,the density threshold was calculated by recursive algorithm,which could solve the shortage of human input and greatly reduce the dependence of the algorithm on prior knowledge.The experimental results show that the algorithm is superior to GP-CLIQUE and CLIQUE in terms of time and clustering accuracy.