Stepwise orthogonalization of predictors in classification and regression techniques: an “old” technique revisited