Table 8.

Calibration performance of depression prediction models.

Model	Model category	ECE ↓	Brier Score ↓
BERT	General-domain transformer	0.092	0.182
RoBERTa	General-domain transformer	0.085	0.176
MentalRoBERTa (Ji et al., 2022)	Mental-health-adapted transformer	0.078	0.169
Prompt-only LLM	Foundation-model baseline	0.121	0.205
Evidence + Logistic Regression	Evidence-based baseline	0.074	0.165
R-GCRED (Calibrated)	Retrieval-grounded calibrated framework	0.038	0.141

↓ indicates that lower values represent better calibration performance. ECE measures the discrepancy between predicted confidence and observed outcome frequency, whereas the Brier Score measures the mean squared difference between predicted probabilities and binary outcomes.