Due to the high collinearity between variables of spectral
data, a principal component analysis (as described
in Alpaydin, 2010) is applied to the training data to obtain
independent variables. Subsequently a linear discriminant
is calculated which is suitable to separate
the training data in class “female” and “male”. Additionally,
for each principal component, the discriminant
strength is determined.
To validate the calculated discriminant, the validation
data is also transformed to principal component
space and classified within it.