Self-harm detection on LRH (held-out test)
0.884AUPRCProposed approach
Evaluation Results
| Method | Links | |
|---|---|---|
| Proposed approachTraining Set=RMH Chunk 3, Input Type=LLM-generated structured evidence, Pipeline=Stage 3 training after Stage 1 screening and Stage 2 evidence extraction2026.06 | 0.884 | |
| ClinicalBERTTraining Set=RMH Chunk 3, Input Type=raw triage note, Pre-screening=full dataset2026.06 | 0.856 | |
| LRTraining Set=RMH Chunk 3, Input Type=raw triage note, Pre-screening=full dataset2026.06 | 0.789 | |
| GBMTraining Set=RMH Chunk 3, Input Type=raw triage note, Pre-screening=full dataset2026.06 | 0.772 |