Table 8 Ablation study.
From: RABEM: risk-adaptive Bayesian ensemble model for fraud detection
Configuration | Accuracy | F1 Score | Recall | Notes |
|---|---|---|---|---|
Full RABEM pipeline | 99.38% | 0.92 | 99.41% | Baseline (all modules included) |
No VAE (raw features only) | 97.46% | 0.76 | 88.23% | Significant drop in recall |
No GRU (no temporal modeling) | 98.11% | 0.82 | 91.34% | Loss in sequential insight |
No GP (no uncertainty modeling) | 98.65% | 0.85 | 93.01% | Higher false positives |
No RPTree (replaced with RF) | 98.24% | 0.81 | 90.45% | Drop in generalization |
No Bayesian fusion (raw classifier output) | 98.91% | 0.88 | 95.10% | Confidence calibration affected |
