💳 Credit Card Fraud Detection System

Internship Project → Production-Ready Machine Learning Application

---

About This Project

This repository contains my Credit Card Fraud Detection project, originally submitted during my CodeClause Data Science Internship (July 2025) and later upgraded into a production-ready, user-friendly machine learning application.

The goal of this project is not just to build a model, but to show how data science is used in the real world from raw data to a system that non-technical users can actually use.

---

Problem Statement

Credit card fraud is rare but extremely costly.
Financial institutions need systems that can:

• Detect suspicious transactions
• Assign a risk score instead of a hard yes/no
• Allow business teams to take decisions like allow, review, or block

This project solves that by predicting a fraud probability score for each transaction and presenting it in a simple, human-understandable dashboard.

---

🗂️ Repository Structure

• CodeClauseInternship_Fraud-Detection/
• │
• ├── internship_submission/
• │ └── credit_card_fraud_detection_basic.ipynb
• │
• |
• ├── production_app/
• │ ├── notebooks/
• |
• │ ├── src/
• │ │ ├── data_loader.py
• │ | ├── evaluate.py
• │ | ├── feature_engineering.py
• │ | ├── preprocessing.py
• │ │ ├── train.py
• │ │ ├── sampling.py
• │ │ └── predict.py
• | |
• │ ├── models/
• │ │ └── fraud_xgb.pkl
• | |
• │ ├── dashboard/
• │ │ └── dashboard.py
• | |
• | ├── api/
• │ │ └── app.py
• | |
• │ ├── run_training.py
• | |
• │ └── requirements.txt
• │
• └── README.md

---

Internship Submission (July 2025)

The original internship work includes:

• Exploratory Data Analysis (EDA)
• Severe class imbalance handling using SMOTE
• Fraud detection using XGBoost
• Hyperparameter tuning with GridSearchCV
• Model evaluation using ROC-AUC
• Model explainability using SHAP

📁 Located inside: internship_submission/

---

Production-Ready Enhancements

After the internship, the project was upgraded to follow industry best practices:

Modular ML Pipeline

• Clear separation of preprocessing, training, and prediction logic

Training–Inference Consistency

• Same preprocessing logic reused during prediction to avoid feature mismatch issues

Interactive Streamlit Dashboard

• CSV upload support
• Fraud probability converted into human-friendly risk levels
• Clear metrics and visualizations
• Downloadable analysis report
• Designed for non-technical users

Real-World Risk Interpretation

Fraud Probability	Risk Level	Suggested Action
< 0.5	Low Risk	Allow transaction
0.5 – 0.7	Medium Risk	Manual review
> 0.7	High Risk	Block / Alert

---

How to Run This Project on Your System

Follow these steps exactly to run the application locally.

---

Clone the Repository

git clone https://github.com/Prince5104/CodeClauseInternship_Fraud-Detection.git
cd CodeClauseInternship_Fraud-Detection

python -m venv venv
source venv/bin/activate   # Linux / macOS
# venv\Scripts\activate    # Windows

pip install -r production_app/requirements.txt

python3 production_app/run_training.py

streamlit run production_app/dashboard/dashboard.py

What the Dashboard Shows

For non-technical users, the dashboard provides:

Fraud probability for each transaction

Clear risk labels:

Low Risk

Medium Risk

High Risk

Summary metrics

Fraud probability distribution graph

Downloadable CSV report

No coding knowledge is required to use it.

If you want to connect with me follow me on:

• Linkedin: www.linkedin.com/in/prince-raj-tech
• Youtube channel: https://www.youtube.com/@SynapseSpaceData