TrustIQ AI: Real-time Bank Fraud Detection with Machine Learning

Highlights

Machine Learning Pipeline

• Dataset: 1M+ synthetic transactions with 100+ features including transaction patterns, user behavior, and merchant risk profiles
• Feature Engineering: Time-based aggregations, velocity checks, geographical features, and behavioral patterns
• Models: XGBoost (primary), Isolation Forest (anomaly detection), Logistic Regression (baseline)
• Imbalance Handling: SMOTE oversampling, class weighting, and ensemble methods
• Evaluation: 99.2% Precision, 98.7% Recall, 98.9% F1-Score, 0.8% False Positive Rate

My Role & Responsibilities

• End-to-end ML pipeline development: data synthesis, feature engineering, model training, and evaluation
• Real-time inference system with FastAPI and WebSocket for instant fraud scoring
• Interactive React dashboard with live transaction monitoring and alert management
• Deployment architecture with Docker, Redis for caching, and PostgreSQL for transaction storage
• Performance optimization for sub-second inference and 99.9% system uptime

Business Impact

• 21.62% fraud rate detection in simulated environment with real transaction patterns
• $2,865+ blocked in fraudulent transactions during testing phase
• Sub-second detection speed enabling real-time transaction blocking
• Reduced false positives by 67% compared to traditional rule-based systems
• Scalable architecture supporting 10,000+ transactions per minute