Abstract 089
Multivariate analysis of routine beer analysis results
J. Amer. Soc. Brew. Chem. 63(3): 113-120, 2005
K.J. Siebert
A study of the collective information content of 47 analytical observations applied to at least six samples each of 10 beer brands was carried out. Principal Components Analysis (PCA) was used to determine the number of fundamental properties represented in the data set and thereby to estimate the degree of redundancy. The number of significant principal components (PCs) is seven or eight, depending on the criterion used. This indicates that the 47 measurements together only contained information on 7 – 8 fundamental properties, and there was considerable redundancy. The first two PCs contained information that was sufficient to almost completely separate samples of the 10 brands. It was possible to identify 14 measurements from the 47 used that captured most of this information and retained much of the ability to separate brands. Hierarchical cluster analysis applied to a correlation matrix of the measurements showed close relationships among some of the measurements. Several pattern recognition procedures were applied to attempt classification of the samples by brand. This was quite successful with Linear Discriminant Analysis (LDA), K-Nearest Neighbor Analysis (KNN) and Soft Independent Modeling of Class Analogy (SIMCA).
Back to K.J. Siebert home page | Back to publication list