Self-harm detection on RMH (prospective)
88.1AUPRCProposed 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 | 88.1 | |
| ClinicalBERTTraining Set=RMH Chunk 3, Input Type=raw triage note, Pre-screening=full dataset2026.06 | 87.4 | |
| GBMTraining Set=RMH Chunk 3, Input Type=raw triage note, Pre-screening=full dataset2026.06 | 83.6 | |
| LRTraining Set=RMH Chunk 3, Input Type=raw triage note, Pre-screening=full dataset2026.06 | 79.2 |