Purpose: To propose a semi-parametric method for conducting scale-invariant sparse principal component analysis (PCA) on high-dimensional non-Caussian data, that is has a weaker modeling assumption and is more robust to possible data contamination compared with sparse PCA. Summary: PCA is an efficient approach for dimension reduction and feature selection. The dimension that is small compared to the sample size can be estimated by eigenvectors of a sample covariance matrix. However when the dimension increases compared to the sample size this method produces poor estimates. To overcome this one approach used is to use sparsity constraints on eigenvectors. Forthis purpose different variants of sparse PCA has been developed and their theoretical properties have been investigated. The drawbacks of PCA and sparse PCA are: They are not scale invariant and are affected when measurement scales are changed, They are not robust to data contaminations or outliers, and They rely on Gaussian or sub-Caussian assumption, which may not be true.
展开▼