27th Annual Student Research Conference
Quantification of Water-Ethanol Mixtures by Principle Component Regression (PCR) and Partial Least-Squares (PLS)
Ai V. Tran*, Benjamin Barndollar, Rebecca Gee, and Uday Malhotra
Dr. Kenneth Carter, Faculty Mentor
Multivariate statistical methods have become increasingly important in analytical chemistry. They allow chemists to extract desirable information from an arbitrarily large number of signals produced by technique such as near-infrared (NIR) spectroscopy. Potential limitations of multivariate analyses, however, include the necessity for appropriate preprocessing, for instance via the Savitzky-Golay smoothing technique, and the pitfall of over-fitting, which fits noise in addition to signal. The computational implementation of multivariate methods is readily available both in proprietary software such as Minitab and free packages in the programming language R. Upon investigating a classic pedagogical article by Beebe and Kowalski, we found that the severity of over-fitting was greater than they stated. In addition, we employ various chemometric methods, such as principle component regression (PCR) and partial least-squares (PLS), in an effort to quantify water-ethanol mixtures by means of NIR spectroscopy.
Keywords: multivariate statistical methods, NIR spectroscopy, Savitzky-Golay smoothing, over-fitting, programming language R, chemometric methods, principle component regression (PCR), partial least-squares (PLS)
Topic(s):Chemistry
Presentation Type: Poster
Session: 5-1
Location: GEO - SUB
Time: 3:30