FoucauldE/doctomodernbert-fr-validator

doctomodernbert-fr-incertain

Binary classifier (SUR / INCERTAIN) fine-tuned from doctolib-lab/doctomodernbert-fr-base.

• Input format: <dp_chap> [SEP] <text>
• Labels: generated by an LLM (Gemini), not human-annotated.
• Training data: synthetic only, no patient data.
• Test metrics (500 examples, INCERTAIN class): precision = 0.870, recall = 0.724, F1 = 0.790
• Accuracy: 0.844 (macro F1 0.833)
• Split: random stratified on dp_chap + label, so codes in the test set also appear in training.
• Limits: not validated on unseen codes; not for clinical decisions.