Feature Scaling for Kernel Fisher Discriminant Analysis Using Leave-One-Out Cross Validation