无监督特征选择的改进稀疏主成分分析算法

An improved sparse principal component analysis algorithm for unsupervised feature selection

为了降低稀疏主成分分析(Sparse Principal Component Analysis,SPCA)算法对高维数据集的计算复杂度,提出一种改进SPCA(Improved Sparse Principal Component Analysis,ISPCA)算法。该算法将特征选择过程分为两个阶段,第一阶段利用不带低秩惩罚项的SPCA先对数据进行一次特征选择,得到降维数据,采用矩阵的广义逆引理降低算法复杂度。第二阶段在降维数据上执行带低秩惩罚项的SPCA对降维数据再次进行特征选择。对比实验结果表明,ISPCA算法比SPCA算法受参数影响较小,特征选择性能更优,运行速度更快。

In order to reduce the computational complexity of sparse principal component analysis(SPCA)algorithm for high dimensional data sets,an improved SPCA(ISPCA)algorithm is proposed.The feature selection process of ISPCA algorithm is divided into two stages.In the first stage,SPCA without low-rank penalty term is used to conduct feature selection on the data to obtain dimension-reduced data.The generalized inverse lemma of matrices is adopted to reduce the complexity of the algorithm.In the second stage,SPCA with low rank penalty term is performed on the dimension-reduced data to conduct another feature selection.The comparative experimental results show that the ISPCA algorithm is less affected by the parameters than the SPCA algorithm.It has better feature selection performance,and runs faster.