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

Area of Science:

• Environmental Chemistry
• Chemometrics
• Data Analysis

Background:

• Spectral data analysis is crucial for understanding aquatic systems.
• Identifying active spectral components requires robust methodologies.
• Existing methods may lack accuracy in determining component numbers.

Purpose of the Study:

• To evaluate Principal Component Analysis (PCA), Principal Object Analysis (POA), and Independent Component Analysis (ICA) for spectral component identification.
• To determine the optimal algorithm for unequivocally identifying active spectral components in aquatic systems.
• To assess the robustness and accuracy of POA and ICA in spectral data analysis.

Main Methods:

• Exploratory data analysis using PCA, POA, and ICA algorithms.
• Application of decision criteria to determine the number of active spectral components.
• Comparative analysis of algorithm performance on spectral datasets from aquatic systems.

Main Results:

• POA demonstrated robustness as an unsupervised, object-oriented exploratory data analysis method.
• POA successfully determined the number of independent spectral components in datasets.
• The combination of POA and ICA proved to be a robust and accurate unsupervised method for retrieving maximal spectral information.

Conclusions:

• POA is a reliable method for determining the number of independent spectral components.
• POA combined with ICA offers a powerful approach for comprehensive spectral information retrieval.
• This combined method accurately identifies component numbers, signal sources, and their contributions in aquatic systems.