外的基準の説明を目的とした独立成分分析

抄録

Independent component analysis (ICA) is an unsupervised technique for signal processing, and is useful for projection pursuit as well. This paper proposes an enhanced technique of ICA, which extracts independent components that are useful for revealing mutual relationship between observations and some external criteria. Fast ICA algorithm is performed after the preprocessing by regression-principal component analysis that extracts latent variables closely related to external criteria from observations. Numerical experiments including knowledge discovery from POS transaction data reveal the characteristic feature of the proposed method.