Credit Card Fraud Detection System

Portfolio Note: Portfolio recreation of production fraud detection system built at Omfys Technologies.

🎯 Overview

Real-time fraud detection validating 500K+ transactions/day with 92% accuracy using Spark Scala, XGBoost, and ONNX with <100ms latency.

📊 Key Metrics

• Volume: 500K+ transactions/day
• Accuracy: 92%
• False Positive Rate: <2%
• Latency: <100ms p95
• Fraud Prevention: ₹10L+ annually

🛠️ Tech Stack

• Processing: Spark (Scala), Kafka Streams
• Storage: HBase, HDFS (Parquet), Hive
• ML: XGBoost, Random Forest, ONNX Runtime
• ETL: Apache Sqoop
• Monitoring: Prometheus, Grafana

⚡ Key Features

1. High-Performance Stream Processing

• Spark Scala for optimal JVM performance
• Exactly-once processing with checkpointing
• Write-ahead logs for fault tolerance

2. Advanced Feature Engineering (50+ features)

• Velocity checks: Transaction frequency patterns
• Geographical anomalies: Location-based risk
• Amount deviation: Statistical outlier detection
• Merchant patterns: Historical analysis
• Device fingerprinting: Behavioral analysis

3. Ensemble ML Models

• Random Forest + XGBoost stacking
• ONNX format for 3x faster inference
• A/B testing framework

4. Low-Latency Profile Store

• HBase for <10ms p95 lookups
• 10M+ transaction profiles
• Bloom filters for efficiency

📁 Project Structure

fraud-detection-system/
├── src/
│   ├── main/scala/          # Spark Scala code
│   │   ├── FraudDetector.scala
│   │   ├── FeatureEngineering.scala
│   │   └── KafkaStreaming.scala
│   └── python/              # ML training
│       ├── model_training.py
│       └── onnx_conversion.py
├── config/
│   ├── hbase-site.xml
│   └── application.conf
├── build.sbt
├── requirements.txt
└── README.md

🚀 Getting Started

git clone https://github.com/Amanroy666/fraud-detection-system.git
cd fraud-detection-system

# Build Scala project
sbt clean compile assembly

# Install Python dependencies  
pip install -r requirements.txt

📈 Performance Results

Metric	Value
Accuracy	92%
Precision	89%
Recall	94%
F1-Score	91.5%
False Positive Rate	1.8%
Latency (p95)	98ms

👤 Author

Aman Roy - Data Engineer at Omfys Technologies
📧 contactaman000@gmail.com | 💼 LinkedIn | 🐙 @Amanroy666