Table 2.

Positive and negative cases in the training, validation, calibration, and test/evaluation data.

Dataset	Partition/experimental role	Depression-positive (n)	Depression-negative/control (n)	Total (N)	Use in R-GCRED
DAIC-WOZ	Official training partition	31	76	107	Source benchmark training data
	Official development/validation partition	12	23	35	Validation and model-development data
	Official test partition	14	33	47	Source benchmark test data
	Complete cohort	57	132	189	Participant-level five-fold CV cohort
	Model-development portion per CV iteration	45–46	105–106	151–152	Model development within five-fold CV
	Held-out evaluation fold per CV iteration	11–12	26–27	37–38	Held-out cross-validation evaluation only
CLEF eRisk 2025 Task 2	Held-out evaluation cohort	102	807	909	Longitudinal early-detection evaluation
	10% timeline observation	102	807	909	Evaluation using first 10% of each user's timeline
	25% timeline observation	102	807	909	Evaluation using first 25% of each user's timeline
	50% timeline observation	102	807	909	Evaluation using first 50% of each user's timeline
	100% timeline observation	102	807	909	Full-timeline evaluation