Table 3.

Cross-validation results per model type within relapse participants listed by median (IQR).

Model	Rank	Days of relative health in train, %	Hidden units	Sensitivity, median (IQR)	Specificity, median (IQR)
FNN ADa	9.28	80	40	0.25 (0.15-1.00)	0.88 (0.14-0.96)
GRU Seq2Seqb	12.72	80	50	0.29 (0.08-0.83)	0.86 (0.24-0.90)

^aFNN AD: fully connected neural network autoencoder.

^bGRU Seq2Seq: gated recurrent unit sequence-to-sequence.