Worker Acceptance Prediction and Utility Estimation on Preisach n=36,891 (full)
0.09NRMSE U1 (Accepted)v1: XGBoost regressors
Evaluation Results
| Method | Links | ||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| v1: XGBoost regressorsUtility estimator=Separate XGBoost on accepted/rejected, Optimal threshold=0.294, P(accept) curves=Flat for most transactions2026.06 | 0.09 | 0.2 | 0.799 | 82.7 | 53.8 | 22,171 | 60.1 | 1,186 | 3.2 | 12,988 | 35.2 | 546 | 1.5 | 31 | 2 | 0.001 | |
| v2: Dual-output NNUtility estimator=Shared encoder + margin loss λ = 0.5, Optimal threshold=0.237, P(accept) curves=Monotone, well-shaped2026.06 | 4.07 | 7.53 | 0.72 | 80.5 | 49.5 | 27,382 | 74.2 | 124 | 0.3 | 9,384 | 25.4 | 1 | 0 | 31 | 7 | 1.452 |