AML alert triage on SynthAML10 AMLSim (test)
0.75PR-AUCRAG + Counterfactual Checks
Evaluation Results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| RAG + Counterfactual ChecksRetrieval=Evidence Retrieval, Counterfactual Checks=Enabled2026.03 | 0.75 | 0.53 | 0.75 | 0.62 | 0.98 | 0.88 | 0.58 | 0.76 | 0.03 | 0.02 | |
| RAGRetrieval=Evidence Retrieval, Counterfactual Checks=None2026.03 | 0.71 | 0.49 | 0.73 | 0.59 | 0.96 | 0.83 | 0.41 | 0.57 | 0.06 | 0.05 | |
| Graph MLModel Type=GNN2026.03 | 0.69 | 0.46 | 0.71 | 0.56 | — | — | — | — | — | — | |
| Tabular MLModel Type=GBDT2026.03 | 0.63 | 0.42 | 0.66 | 0.51 | — | — | — | — | — | — | |
| LLM-onlyRetrieval=None2026.03 | 0.52 | 0.31 | 0.6 | 0.41 | 0.62 | 0.54 | 0.28 | 0.39 | 0.11 | 0.14 | |
| Rule/Threshold Baseline2026.03 | 0.41 | 0.18 | 0.78 | 0.29 | — | — | — | — | — | — |