Fig. 5.
Prediction uncertainty when using train/test data splitting for validation. Receiver Operator Characteristic curves for training/test performance of PLS-DA on data set ST001047 for five iterations of stratified random splitting (2/3 training and 1/3 test). Green line = ROCtrain, yellow line = ROCtest. This resulted in: a AUCtrain = 0.96, AUCtest = 0.87; b AUCtrain = 0.97, AUCtest = 0.96; c AUCtrain = 0.97, AUCtest = 0.88; d AUCtrain = 0.98, AUCtest = 0.90; e AUCtrain = 0.99, AUCtest = 0.98. d The 95% OOB confidence interval for the same data. Note all a–e ROCtest curves lie within the 95% confidence interval