Table 3 Summary of final datasets: train and test
From: The effects of class rarity on the evaluation of supervised healthcare fraud detection models
| Â | Dataset | Features | Non-fraud | Fraud | % Fraud |
|---|---|---|---|---|---|
Train | Part B | 126 | 3,691,146 | 1409 | 0.038 |
Part D | 126 | 2,098,715 | 1018 | 0.048 | |
DMEPOS | 145 | 862,792 | 635 | 0.074 | |
Combined | 173 | 759,267 | 473 | 0.062 | |
Test | Part B | 126 | 999,815 | 99 | 0.010 |
Part D | 123 | 744,918 | 135 | 0.018 | |
DMEPOS | 119 | 290,548 | 75 | 0.026 | |
Combined | 171 | 256,529 | 55 | 0.021 |