Table 4 Mean and standard deviation \(\sigma\) of test \(F_1\) scores across 10 models trained using best-performing (\(F_1\) on validation dataset) configuration found in 30 trials of hyperparameter optimization. Numbers rounded to three decimal places, best configuration of each disease marked bold
Type 2 diabetes | Glaucoma | ||
|---|---|---|---|
Model | Mean \(F_1\) (\(\pm \sigma\)) | Model | Mean \(F_1\) (\(\pm \sigma\)) |
Extractive | |||
 flan-t5-base | 0.547 (± 0.006) | flan-t5-base | 0.636 (± 0.006) |
 led-base-16384 | 0.525 (± 0.009) | led-base-16384 | 0.572 (± 0.010) |
 longformer-base-4096 | 0.540 (± 0.008) | longformer-base-4096 | 0.613 (± 0.007) |
Generative | |||
 flan-t5-base | 0.539 (± 0.029) | flan-t5-base | 0.584 (± 0.025) |
 led-base-16384 | 0.400 (± 0.079) | led-base-16384 | 0.353 (± 0.106) |