Development and validation of a meta-learner for combining statistical and machine learning prediction models in individuals with depression

Table 2 Evaluation on post treatment Patient Health Questionnaire (PHQ-9) and all-cause dropout predictions made by the statistical, machine learning base-learners and meta-learners

	PHQ-9 score MAE^* [95% CI^†]	Dropout AUC^‡ [95% CI]
Base-learner ridge regression	4.64 [4.57, 4.71]	0.597 [0.594, 0.601]
Base-learner MLP^§	4.63 [4.57, 4.71]	0.598 [0.594, 0.601]
Meta-learner linear/logistic regression	4.54 [4.48, 4.60]	0.604 [0.600, 0.606]
Meta-learner MLP	4.52 [4.45, 4.59]	0.604 [0.600, 0.606]

^*Mean absolute error.
^†Confidence interval, calculated as the 2.5th to the 97.5th percentile of bootstrap estimates.
^‡Area under the receiver operating characteristic curve.
^§Multi-layer perceptron.

ISSN: 1471-244X