Independent Component Analysis with Preprocessing Based on Regression-Principal Component Analysis

Abstract

Independent component analysis (ICA) is an unsupervised technique, which uses higher order statistics than principal component analysis (PCA) to reveal the intrinsic structure of data sets, and is useful for projection pursuit as well. However, in real applications, it is often the case that we fail to extract useful latent variables because they have no connection with predefined criterion variables. This paper proposes an enhanced technique of ICA, which extracts independent components closely related to some external criteria. Preprocessing is performed by using regression-principal component analysis, which estimates latent variables that have high correlation with the external criteria.