Table 4.1:

Synthetic data test set metrics (mean ± std. dev. across 10 experiments) and qualitative assessment of successful ground truth cluster recovery across experiments.

Model	AUROC	AUPRC	Silhouette Index	Cluster Prediction Accuracy	Recovery
Logistic Regression	0.90±0.01	0.90±0.00	0.09±0.01	0.60±0.06	✗
LSTM	0.95±0.01	0.95±0.00	0.09±0.01	0.60±0.05	✗
RETAIN	0.95±0.01	0.95±0.00	0.33±0.06	0.69±0.06	❈
DIPOLE	0.95±0.01	0.95±0.00	0.30±0.06	0.69±0.04	❈
Transformer:BERT	0.95±0.01	0.95±0.00	0.11±0.01	0.67±0.03	✗
AC-TPC	0.95±0.01	0.95±0.00	0.09±0.01	0.77±0.03	✗
T-Phenotype	0.95±0.01	0.95±0.00	0.49±0.32	0.81±0.03	❈
SMD-SSL + SimCLR (Frozen)	0.84±0.01	0.83±0.00	0.15±0.05	0.61±0.02	❈
SMD-SSL + VICReg (Frozen)	0.79±0.01	0.79±0.00	0.12±0.05	0.60±0.03	❈
SMD-SSL + SimCLR (Fine-tune)	0.94±0.01	0.94±0.00	0.39±0.10	0.78±0.04	❈
SMD-SSL + VICReg (Fine-tune)	0.95±0.01	0.95±0.00	0.24±0.12	0.73±0.04	❈
Simple-SCL / Temporal-SCL (PT:✗, NN:✗, TR:✗)	0.93±0.01	0.94±0.01	0.19±0.01	0.64±0.06	✗
Temporal-SCL (PT:✓, NN:✗, TR:✗)	0.94±0.01	0.94±0.00	0.24±0.01	0.62±0.06	✗
Temporal-SCL (PT:✗, NN:✓, TR:✗)	0.94±0.01	0.95±0.00	0.82±0.01	0.84±0.05	❈
Temporal-SCL (PT:✗, NN:✗, TR:✓)	0.95±0.01	0.95±0.00	0.40±0.01	0.57±0.04	✗
Temporal-SCL (PT:✗, NN:✓, TR:✓)	0.95±0.01	0.94±0.00	0.86±0.00	0.91±0.06	❈
Temporal-SCL (PT:✓, NN:✗, TR:✓)	0.95±0.01	0.95±0.00	0.55±0.01	0.56±0.05	✗
Temporal-SCL (PT:✓, NN:✓, TR:✗)	0.95±0.01	0.95±0.00	0.92±0.01	0.85±0.03	❈
Temporal-SCL (Full)	0.95±0.01	0.95±0.00	0.98±0.00	1.00±0.00	✓

PT: Pre-Training, NN: Nearest Neighbor pairing, TR: Temporal Regularization

✗: No correct recovery of the embedding structure.

❈: Partial clustering structure found, but not a completely correct recovery of the embedding structure.

✓ : Completely correct recovery of the embedding structure.