摘要
局部异常挖掘是大规模数据集正常使用过程中不可缺少的步骤,但异常挖掘过程易受冗余数据的干扰。为解决上述问题,提出基于堆栈模型的大规模数据集中局部异常挖掘方法。采用离散小波变换剔除大规模数据中存在的噪声,避免挖掘过程受到噪声干扰,通过收缩因子与移动因子的离散化计算,降低因小波变换引起的数据冗余度,将降噪后的数据输入到堆栈模型中的自编码器中完成特征提取。采用独立成分分析算法对提取的特征实行投影计算,检测出数据中的异常值,完成大规模数据集中局部异常挖掘。实验结果表明,所提方法的挖掘时间短、挖掘准确率高、挖掘精度高。
Local anomaly mining is an indispensable step in the use of large-scale data sets.However,the process of anomaly mining is easily disturbed by redundant data.Therefore,a method of mining local anomalies in large-scale data sets was proposed based on stack model.Firstly,discrete wavelet transform was adopted to eliminate the noise from large-scale data,thus avoiding noise interference.Based on the discretization calculation of the con-traction factor and movement factor,the data redundancy caused by the wavelet transform was reduced.And then,the denoised data was input into the autoencoder in the stack model,thus completing the feature extraction.After that,the independent component analysis algorithm was used to perform the projection calculation on the features.Finally,out-liers in the data were detected to complete local anomaly mining.Experimental results show that the proposed method has short mining time and high mining accuracy.
作者
陈滢生
周宪章
CHEN Ying-sheng;ZHOU Xian-zhang(School of Computer Engineering,Chongqing College of Humanities,Science and Technology,Chongqing 401524,China;School of Computer and Information Science,Southwest University,Chongqing 400715,China;Chongqing Institute of Educational Sciences,Chongqing 400015,China)
出处
《计算机仿真》
北大核心
2023年第6期495-498,508,共5页
Computer Simulation
基金
重庆市教委科学技术研究项目(KJQN202001803)
重庆市教委科学技术研究计划重大项目(KJZD-M202114401)
重庆人文科技学院科学研究项目(CQRKZK202005)。
关键词
离散小波变换
自编码器
独立成分分析
分离矩阵
主成分分析
Discrete wavelet transform
Autoencoder
Activation function
Independent component analysis
Separation matrix
Principal component analysis
