3.2. Linear discriminant analysis
After a stepwise PCA using the more discriminating vari- ables, a linear discriminant analysis was run in order to op- timise the separation of the musts under study and in order to find an operative classification role for discriminating the four must varieties that make Madeira wine. Fig. 6 shows a projection of the musts in 2-D space, explaining 98.4% of the total variance. Four groups representing each variety were clearly observed. The first two discriminant functions (roots) were effective in discriminating between must vari- eties.
The classification capacity of the functions obtained was evaluated introducing ungrouped samples in the initial ma- trix. Hundred percent of the objects (8/8) were correctly