Principal component analysis (PCA) has well-documented merits for dataextraction and dimensionality reduction. PCA deals with a single dataset at atime, and it is challenged when it comes to analyzing multiple datasets. Yet incertain setups, one wishes to extract the most significant information of onedataset relative to other datasets. Specifically, the interest may be onidentifying, namely extracting features that are specific to a single targetdataset but not the others. This paper develops a novel approach for suchso-termed discriminative data analysis, and establishes its optimality in theleast-squares (LS) sense under suitable data modeling assumptions. Thecriterion reveals linear combinations of variables by maximizing the ratio ofthe variance of the target data to that of the remainders. The novel approachsolves a generalized eigenvalue problem by performing SVD just once. Numericaltests using synthetic and real datasets showcase the merits of the proposedapproach relative to its competing alternatives.
展开▼
