摘要
为了提高基于一范数的核主成分分析算法(KPCA-L1)处理异常检测问题的速度,提出了基于样本选取和加权KPCA-L1的异常检测方法。该方法从训练集中选取具有代表性的特征子集,然后为所得特征子集中的样本赋予权重,用带有权重的特征子集训练模型,构造加权KPCA-L1。与KPCA-L1相比,该方法能够有效地减小训练集的规模,同时改善了KPCA-L1算法的更新方法。在人工数据集和标准数据集上的实验结果表明,在保证异常检测准确率的前提下,该方法比KPCA-L1具有更快的建模速度。
To enhance the speed of L1 norm based KPCA（ KPCA-L1） for tackling novelty detection problems,this paper proposed a novelty detection method based on sample selection and weighted KPCA-L1. For the proposed method,it selected the representative feature subset from the given training set firstly. Furthermore,it signed the samples in the obtained feature subset with weights and used such feature subset to construct the weighted KPCA-L1. In comparison with KPCA-L1,the proposed method can efficiently reduce the size of training set and improve the update way of KPCA-L1. Experimental results on the synthetic and benchmark data sets demonstrate that,compared to KPCA-L1,the proposed method can obtain faster modeling speed on the premise of assuming the accuracy rate of novelty detection.
出处
《计算机应用研究》
CSCD
北大核心
2016年第5期1354-1358,共5页
Application Research of Computers
基金
国家自然科学基金资助项目(60903089
61473111)
河北省自然科学基金资助项目(F2013201060)
关键词
核主成分分析
一范数
样本选取
异常检测
KPCA（kernel principal component analysis）
L1 norm
sample selection
novelty detection
