Improving principal component analysis using Bayesian estimation

Bayesian estimation is used in this paper to derive a new PCA (principal component analysis) modeling algorithm that improves the estimation accuracy by incorporating prior knowledge about the data and model. It is shown that the algorithm is more general than the existing methods [PCA and MLPCA (maximum-likelihood PCA)], and reduces to these techniques when a uniform prior is used. It is also shown that, when no external information is available, an empirically estimated prior from the available data can still provide improved accuracy over non-Bayesian methods.