Linear discriminant analysis attemots to create one or more linear functions to separate known classes. In the case of two classes and 2D data this is simply a line that runs between the classes. Ideally the classes are completely separated by the linear functions, but in soem case this may be impossible, either becasue of noise blurring the boundaries (in which case the linear discrimant may still be a good classifier), or becasue the classes are not linearly separable.
Used on pages 125, 135
Also known as linear discriminant
Linearly separable clusters