Data fitted the selected model relatively well when the modeling approach was conducted by region (training and testing AUCs < 0.75).
All models, except the CART model, performed well when the reference model RF was used (AUC > 0.8, Table 4 .Considering the learning curves, the CART models performed poorly.
The first perspective is that a good learning algorithm should perform well when trained and applied to a single population and experimental setting, but it is not expected to perform well when the resulting model is applied to different populations and settings.
REML performed well when the random effects model is misspecified using a t-distribution; Ma and Mazumdar 2011 found that this was also the case for other random effects distributions.
This approach was shown to perform well when the target signal matches the model (signals containing only target drums) [11].
The test showed that the model was well calibrated with P = .172 indicating that there is no large discrepancy between observed and expected results and indicating also that the data fit the model well when the clinical hours were used as a predictor for hand and finger discomfort.
Such algorithms work well when the EDA's statistical model matches the structure of the domain.
The model calibrates well when the calibration curve is close to the bisector.
