Table 3 Performance comparison of models with random initialization versus transfer learning initialization (pre-trained on KoGES clinical data) on the SNUH dataset.
From: Transfer learning prediction of type 2 diabetes with unpaired clinical and genetic data
Initialization | Accuracy | Sensitivity | Specificity | AUC | Max AUC |
|---|---|---|---|---|---|
Random (scratch) | 0.8234 | 0.7832 | 0.8274 | 0.8952 | 0.9549 |
Pre-trained (KoGES) | 0.8205 | 0.7796 | 0.8246 | 0.8971 | 0.9552 |
