Skip to content

README.md

Latest commit

 

History

History
84 lines (55 loc) · 2.64 KB

README.md

File metadata and controls

84 lines (55 loc) · 2.64 KB
Raw

🛡️ Real-Time Fraud Detection System

Fraud Detection Architecture

📖 Overview

This project is a real-time fraud detection system built using a modern data pipeline architecture. It integrates multiple tools for data ingestion, processing, modeling, and analysis, orchestrated with Docker and Apache Airflow.

🧱 System Components

🏦 System Fraud Layer

  • Bank System: Simulated transaction data source.
  • Kafka: Real-time streaming platform to capture and forward transaction data.
  • Apache Spark: Performs real-time processing and scoring of streaming data.
  • Gmail API: Sends alert notifications when fraudulent activity is detected.

🧠 Modeling Layer

  • MLflow: Used for model tracking, versioning, and experiment logging.
  • Training Service: Responsible for training fraud detection models.
  • MinIO: Object storage to save trained models (S3 compatible).
  • PostgreSQL: Stores structured training data and metadata.

⚙️ Orchestration Layer

  • Apache Airflow: Manages DAGs for ETL, training, and monitoring pipelines.
  • PostgreSQL: Backend metadata database for Airflow.
  • Celery + Redis: Used for distributed task queuing and execution in Airflow.

📊 Analysis Layer

  • Google BigQuery: Centralized data warehouse for transaction logs and model results.
  • Power BI: Business Intelligence dashboard for fraud trend analysis and reporting.

🐳 Dockerized Services

All services are containerized with Docker for reproducibility and scalability:

  • airflow-webserver, airflow-scheduler, airflow-worker
  • mlflow-server
  • kafka, zookeeper
  • postgres, redis, minio
  • spark, bigquery-connector

🔄 Data Flow Summary

  1. Simulated transactions are sent to Kafka.
  2. Spark consumes the stream, processes it, and applies the fraud detection model.
  3. Detected frauds trigger email alerts and are logged into BigQuery.
  4. Airflow orchestrates batch jobs like retraining models or data ingestion.
  5. MLflow tracks experiment metadata and trained model performance.
  6. Power BI visualizes fraud statistics via BigQuery integration.

🚀 Getting Started

# Clone repository
git clone https://github.com/your-username/fraud-detection-realtime.git
cd fraud-detection-realtime

# Start the system
docker-compose up --build

Ensure .env and credential files are properly configured (see .env.example).

⚠️ Security Note

🚫 Do not commit cloud credentials (e.g. GCP, AWS, service account .json) to this repository.