Fraud Detection System
Real-Time Fraud Detection
System
XGBoost + Isolation Forest ensemble trained on 284,807 credit card transactions. Achieves 97.4% ROC-AUC with SHAP explainability.
97.4%
ROC-AUC
86.3%
Avg Precision
82.1%
F1 Score
284K
Transactions
0.17%
Fraud Rate
⚡ Live Transaction Simulator
Adjust sliders → see prediction update
ANALYSING...
—
FRAUD PROBABILITY
XGBoost
—
Anomaly
—
📊 Feature Importance (SHAP)
🎯 Confusion Matrix
Test set: 56,962 transactions
381
True Positives
Fraud caught ✓
1,204
False Positives
False alarms
49
False Negatives
Missed fraud ✗
55,328
True Negatives
Correct legit ✓
BUSINESS IMPACT
$45,720 fraud prevented
$5,880 missed (49 × avg $120)
🏗️ Model Architecture
INPUT
Raw Transaction
30 features
→
ENGINEER
Feature Eng.
+7 features
→
SMOTE
Balance Data
0.17% → 50%
→
XGBoost (75%)
Supervised
Isolation Forest (25%)
Anomaly Detection
→
OUTPUT
Fraud Score
+ SHAP explain