Universum Principal Component Analysis

In this paper, a novel dimensionality reduction method is developed based on both the principal component analysis (PCA) and Universum learning. PCA works on the target data, ignoring the Universum-do not belong to either class of interest, may contain useful prior knowledge in the same domain as the problem at hand, which has been proved to be helpful in classification and clustering. The proposed method projects target data and Universum into two orthogonal complement spaces with the aim of minimizing the reconstruct error respectively, thus named as universum principal component analysis (UPCA). Experimental results on the UCI datasets and USPS datasets show its effectiveness compared to traditional PCA.