Surface analysis insight note. Principal component analysis (PCA) of an X‐ray photoelectron spectroscopy image. The importance of preprocessing

Abstract

This Insight Note follows two previous Insight Notes on XPS image analysis that dealt with the importance of analyzing the raw data and the use of summary statistics. As a next step in the exploratory data analysis (EDA) of XPS images, we now show principal component analysis (PCA) of an XPS image. PCA is appropriate when the spectra in a data set are correlated to some degree and the noise in the spectra is unimportant. In these cases, PCA can significantly reduce the dimensionality and complexity of data sets. Preprocessing is an important part of many PCAs. Its usefulness is illustrated with a small, mock data set, where the potential pitfalls of not preprocessing are shown. PCAs of an XPS image data that was not preprocessed and preprocessed by mean centering are illustrated. Scree plots, which are used to determine the number of abstract factors (principal components, PCs) that describe a data set, are shown. The spectra in our XPS image are quite noisy, which is consistent with the moderate, but still significant, amount of variance that is captured by the first two PCs in our PCA. With both preprocessing methods, the loadings on PC1 and PC2 are remarkably smooth. The loadings on the next six PCs also appear to contain some chemical information. Scores images generated using both no preprocessing and preprocessing by mean centering reveal many of the same general features in the data set that were found in our two previous Insight Notes.

Surface analysis insight note: An example of a cluster analysis of spectra from an X‐ray photoelectron spectroscopy image

“…This insight note is the fifth in a series that describes different chemometrics/informatics approaches for analyzing XPS image data sets. The previous notes focused on (i) plotting the data in different ways, including examining slices through the data cube and looking at average and standard deviation spectra 6 ; (ii) summary statistics, especially pattern recognition entropy (PRE) 7 ; (iii) principal component analysis (PCA) 8 which is often the first exploratory data analysis (EDA) method used on a dataset, 9–14 ; and (iv) multivariate curve resolution (MCR), 15 which enjoys various advantages over PCA because of its non‐negativity constraints and the fact that its factors do not need to be orthogonal 16–18 . Other researchers have similarly recognized the importance of chemometrics/informatics methods for analyzing large XPS data sets 19–24 …”