Fabricated Table Detection on FABTAB GPT-4o In-Domain (test)
0.987AUROCFusedAll (RandomForest)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| FusedAll (RandomForest)Observer=GPT-22026.03 | 0.987 | 0.985 | 0.933 | 0.054 | |
| FusedAll (LogReg)Observer=GPT-22026.03 | 0.986 | 0.971 | 0.856 | 0.03 | |
| FusedAll (MLP)Observer=GPT-22026.03 | 0.986 | 0.975 | 0.862 | 0.03 | |
| TAB-AUDIT (PPL4, LogReg)Observer=Qwen2026.03 | 0.902 | 0.855 | 0.354 | 0.054 | |
| TAB-AUDIT (PPL4, LogReg)Observer=GPT-22026.03 | 0.884 | 0.804 | 0.374 | 0.074 | |
| TAB-AUDIT (PPL4, LogReg)Observer=Llama2026.03 | 0.866 | 0.82 | 0.246 | 0.039 | |
| Binoculars (no training)Observer=shared2026.03 | 0.683 | 0.701 | 0.385 | 0.108 | |
| DetectGPT-style (no training)Observer=GPT-22026.03 | 0.397 | 0.43 | 0.015 | 0.02 |