Pharmaceutical Statistics Using SAS

2.5. Partial Least Squares for Discrimination

Perhaps the most commonly used method of statistical discrimination is simple linear discriminant analysis (LDA). A fundamental problem associated with the use of LDA in practice, however, is that the associated pooled within-groups sums-of-squares and cross-products matrix has to be invertible. For data sets with collinear features or with significantly more features than observations this may not be the case. In these situationsthere are many different options available. By far the most common in the practice of chemometrics is to first use principal components analysis (PCA) to reduce the dimension of the data, and then to follow the PCA with a version of LDA, or to simply impose an ad hoc classification ...