Real-Time Fraud Detection System | MSc Data Science Project

📌 Project Overview

This project implements a production-grade real-time fraud detection system designed to identify fraudulent financial transactions using machine learning and MLOps best practices.

Unlike academic-only ML projects, this system focuses on the entire lifecycle: data ingestion, preprocessing, model training, deployment, automation, and monitoring readiness.

🎯 Objectives

Detect fraudulent transactions with high precision
Handle large-scale, highly imbalanced datasets
Separate experimentation from production code
Deploy ML models as real-time REST APIs
Automate ML workflows using CI/CD

📊 Dataset

Source: Kaggle Credit Card Transaction Dataset
Files: fraudTrain.csv, fraudTest.csv
Target Variable: is_fraud
Key Challenge: Extreme class imbalance

🔬 Exploratory Data Analysis (EDA)

Fraud vs non-fraud distribution analysis
Transaction amount distribution & outliers
Log-scaled visualizations for skewed data
Feature relevance exploration

🧠 Feature Engineering & Modeling

Encoding of high-cardinality categorical variables
Numerical feature scaling
Class imbalance handling using weighted models
Random Forest selected for robustness & interpretability

🌐 Real-Time Prediction API

Flask-based REST API
Health endpoint for service monitoring
Prediction endpoint for fraud probability
Model loaded dynamically at runtime

🐳 Docker & Deployment

Fully containerized application
Environment consistency across systems
Production-ready Dockerfile

🤖 CI/CD with Jenkins

Automated code checkout from GitHub
ETL pipeline execution
Model training & evaluation
Docker image build & deployment

🛠️ Technology Stack

Python Pandas Scikit-learn Flask Docker Jenkins GitHub Jupyter Notebook

🎓 Academic & Industry Relevance

This project was developed as part of an MSc Data Science capstone and demonstrates strong alignment with industry ML engineering and MLOps practices.

It emphasizes building ML systems, not just ML models.

🚀 Future Enhancements

Real-time streaming with Kafka
Model drift detection & monitoring
Automated retraining pipelines
Cloud deployment (AWS / GCP)