基于谱聚类的多维数据集异常数据检测方法

Detection method of abnormal data in cube based on spectral clustering

针对多维数据集异常数据检测过程中未对多维数据集进行降维处理,导致多维数据集中异常数据检测精度较低、误检率较高、检测时间较长的问题,提出一种基于谱聚类的多维数据集异常数据检测方法。首先,通过拉普拉斯矩阵对多维数据集中的数据进行聚类,初步分类数据;其次,采用局部线性嵌入(LLE)算法对分类后的数据进行降维处理,用特征向量表达高维数据集,去除多维数据集中的冗余信息;最后,将处理后的多维数据集输入到支持向量机模型中,根据回归估计值的计算,完成异常数据的检测。实验结果表明,本文算法进行多维数据集中异常数据检测的精度更高、误检率更低,检测时间较短。

Due to the lack of dimension reduction in the process of cube abnormal data detection,the detection accuracy of abnormal data in the cube is low,the error detection rate is high,and the detection time is long.Therefore,a cube abnormal data detection method based on spectral clustering is proposed.Cluster the data in the multidimensional data set through Laplace matrix,preliminarily classify the data,use LLE algorithm to reduce the dimension of the classified data,express the high-dimensional data set with eigenvectors,remove the redundant information in the multidimensional data set,input the processed multidimensional data set into the support vector machine model,and complete the detection of abnormal data according to the calculation of regression estimates.Experimental results show that the proposed algorithm has higher accuracy,lower false detection rate and shorter detection time.