Table 4.

Depression classification performance on the DAIC-WOZ dataset.

Model	Model category	Macro-F1 ↑	AUC ↑	Sensitivity ↑
BERT Fine-tuned	General-domain transformer	0.74	0.80	0.71
RoBERTa Fine-tuned	General-domain transformer	0.76	0.82	0.73
MentalRoBERTa (Ji et al., 2022)	Mental-health-adapted transformer	0.80	0.86	0.78
Prompt-only LLM	Foundation-model baseline	0.78	0.83	0.75
Evidence + Logistic Regression	Evidence-based baseline	0.79	0.84	0.76
R-GCRED (Proposed)	Retrieval-grounded calibrated framework	0.84	0.89	0.82

↑ indicates that higher values represent better performance.