Model Evaluation Metrics
Raw accuracy alone hid a clinically meaningful precision-recall trade-off. Evaluation included accuracy, balanced accuracy, precision, recall, F1-score, and ROC-AUC. The paper contains an inconsistency about whether the stacking ensemble o…
1 sources - 5 claims
Raw accuracy alone hid a clinically meaningful precision-recall trade-off. Evaluation included accuracy, balanced accuracy, precision, recall, F1-score, and ROC-AUC. The paper contains an inconsistency about whether the stacking ensemble outperformed the individual classifiers. The paper argues that missed diabetic cases make recall clinically important in diabetes screening. The main practical implication is that diabetes screening models should prioritize recall and ROC-AUC over accuracy.