一种基于网格框架下改进的多密度SNN聚类算法

AN IMPROVED MULTI-DENSITY SNN CLUSTERING ALGORITHMBASED ON GRID FRAMEWORK

针对数据集中数据分布密度不均匀以及存在噪声点,噪声点容易导致样本聚类时产生较大的偏差问题,提出一种基于网络框架下改进的多密度SNN聚类算法。网格化递归划分数据空间成密度不同的网格,对高密度网格单元作为类簇中心,利用网格相对密度差检测出在簇边界网格中包含噪声点;使用改进的SNN聚类算法计算边界网格内样本数据点的局部密度,通过数据密度特征分布对噪声点进行类簇分配,从而提高聚类算法的鲁棒性。在UCI高维的数据集上的实验结果表明,与传统的算法相比,该算法通过网格划分数据空间和局部密度峰值进行样本类簇分配,有效地平衡聚类效果和时间性能。

Aiming at the problem of uneven data distribution density in the data set and the existence of noise points,which can easily cause large deviations in the clustering of samples,we propose an improved multi-density SNN clustering algorithm based on the network framework.The data space was recursively divided into grids with different densities.For high-density grid cells as cluster centers,the relative density difference of the grid was used to detect the inclusion of noise points in the cluster boundary grid.The improved SNN clustering algorithm was used to calculate the local density of sample data points in the boundary grid,and the noise points were clustered through the characteristic distribution of data density,so as to improve the robustness of the clustering algorithm.The experimental results on UCI high-dimensional data sets show that,compared with the traditional algorithm,the proposed algorithm can effectively balance the clustering effect and time performance by grid dividing the data space and local density peak for sample cluster allocation.