Table 3.

Fold-specific distribution of depression-positive and depression-negative cases used for training, validation, and calibration in DAIC-WOZ.

Cross-validation iteration	Training positive	Training negative	Validation positive	Validation negative	Calibration positive*	Calibration negative*
Fold 1	39	92	6	14	6	14
Fold 2	39	92	6	14	6	14
Fold 3	40	91	6	14	6	14
Fold 4	40	91	6	14	6	14
Fold 5	40	92	6	14	6	14

^*Positive and negative denote depression-positive and depression-negative participants, respectively. The validation/calibration observations are a subset of the model-development portion reported in Table 2. Calibration does not represent an additional or overlapping independent partition: post-hoc probability calibration was performed using the corresponding validation subset.