Preprocessing Matters: Visualizing the Effect of Spectral Preprocessing on Principal Component Analysis

Abstract

Principal component analysis (PCA) is widely used in chemistry for the exploratory analysis of multivariate data sets; however, many learners have difficulty understanding how spectral preprocessing influences PCA interpretation. This activity was developed to illustrate the effects of preprocessing on spectral profiles and PCA score plots using guided computational activities implemented in R. During four 50 min class periods, participants compared PCA models generated from raw spectra, Savitzky–Golay (SG) smoothing, first- and second-derivative preprocessing, Standard Normal Variate (SNV) normalization, and combinations of these preprocessing methods (Anal. Chem. 1964, 36, 1627.). Guided questions emphasized how preprocessing modifies spectral appearance, sample clustering, intergroup distances, and the proportion of variance explained by the principal components. An assessment using an independent Raman spectroscopy data set evaluated the transfer of these concepts to previously unseen data. Comparison of the different preprocessing strategies demonstrated that smoothing and derivative preprocessing improved spectral discrimination by reducing baseline variation and emphasizing chemically relevant features, whereas SNV normalization minimized intensity differences among spectra. The activities also demonstrated that preprocessing substantially alters the variance structure of spectral data sets, leading to changes in cluster separation and PCA interpretation. By directly relating spectral transformations to multivariate results, this activity provides an accessible instructional activity for teaching spectral preprocessing and principal component analysis in analytical chemistry, spectroscopy, and chemometrics courses.