Restatement Prediction on Audit Analytics Non-Reliance Restatement 2023 (test)
78.2AUROCUMAR
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| UMARmode=Ours2026.06 | 78.2 | 34.1 | 0.068 | 0.052 | |
| Dual-Agent LLMconfig=following Lu et al.2026.06 | 74.1 | 29.5 | 0.074 | 0.079 | |
| XGBoostrounds=300, learning rate=0.05, max depth=62026.06 | 73.8 | 28.7 | 0.076 | 0.083 | |
| FinBERTmode=fine-tuned on MD&A text, epochs=3, batch size=16, learning rate=2x10^-5, max tokens=5122026.06 | 72.1 | 27.1 | 0.079 | 0.089 | |
| Random Foresttrees=500, max depth=82026.06 | 71.3 | 25.9 | 0.081 | 0.094 | |
| Longformermax tokens=4,096, epochs=32026.06 | 70.9 | 26.2 | 0.082 | 0.095 | |
| Single-LLMbackbone=GPT-4o, temperature=0, max output tokens=1,0242026.06 | 69.5 | 23.7 | 0.085 | 0.108 | |
| Logistic Regressionfeatures=all features2026.06 | 67.2 | 21.4 | 0.089 | 0.112 |