🏦Banking Fraud Models·Simulation dataset: 100,000 banking transactions · 0.5% fraud prevalence
Multi-Model Threshold Tuning
Adjust detection thresholds for each AI model. The simulator calculates real-time precision, recall, and F1 metrics on a synthetic dataset of 100,000 transactions with 0.5% fraud prevalence.
Quick Presets
Detection Model Thresholds
LSTM Velocity Engine
Transaction velocity & sequence anomaly detection
65
0 — LENIENT100 — STRICT
GNN Similarity Engine
Graph neural network entity relationship scoring
70
0 — LENIENT100 — STRICT
Isolation Forest
Unsupervised multivariate anomaly isolation
65
0 — LENIENT100 — STRICT
Behavioral Deviation
Biometric & behavioral pattern deviation score
60
0 — LENIENT100 — STRICT
Network Risk Propagation
Risk score propagation through entity network
65
0 — LENIENT100 — STRICT
Flags Generated
5,675
True Positives
270
False Positives
5,405
Missed Fraud
230
8.7%
F1
Precision4.8%
Recall54%
Catch Rate54%
FP Rate5.43%
✗ SUB-OPTIMAL — RETUNING RECOMMENDED
Configuration generates excessive false positives (5,405) or misses too many fraud cases (230). Try the "Balanced" preset.