Unsupervised component analysis: PCA, POA and ICA data exploring - connecting the dots

• Spectral deconvolution of synchronous fluorescence spectra of complex systems
• POA and PCA were used in order to retrieve the number of independent components.
• ICA, PCA and POA are able to deconvolute spectral information.
• POA is very robust in detecting the number of independent components.
• POA-ICA is suggested as the best processing tool for robust unsupervised spectral analysis.

Under controlled conditions, each compound presents a specific spectral activity. Based on this assumption, this article discusses Principal Component Analysis (PCA), Principal Object Analysis (POA) and Independent Component Analysis (ICA) algorithms and some decision criteria in order to obtain unequivocal information on the number of active spectral components present in a certain aquatic system.The POA algorithm was shown to be a very robust unsupervised object-oriented exploratory data analysis, proven to be successful in correctly determining the number of independent components present in a given spectral dataset.In this work we found that POA combined with ICA is a robust and accurate unsupervised method to retrieve maximal spectral information (the number of components, respective signal sources and their contributions).

Figure optionsDownload as PowerPoint slide