🕵️ Fraud Detection ML Project

End-to-End Machine Learning System for Financial Fraud Detection
Achieved 94% Detection Accuracy with Random Forest & XGBoost

---

📌 About the Project

A complete end-to-end Machine Learning pipeline for detecting fraudulent financial transactions. The system handles highly imbalanced datasets, trains and compares multiple ML models, and deploys a real-time prediction interface using Streamlit — all in Python.

This project simulates a real-world fraud detection system used by banks and fintech companies to protect customers from unauthorized transactions.

---

🎯 Business Problem

Financial fraud costs billions of dollars globally every year. Traditional rule-based systems miss sophisticated fraud patterns. This ML-powered system:

• Automatically learns fraud patterns from historical data
• Detects fraudulent transactions in real-time
• Handles heavily imbalanced datasets (fraud is rare — <1% of transactions)
• Provides interpretable results for business decision-making

---

✨ Features

• ✅ End-to-end ML Pipeline — Data → Preprocessing → Training → Evaluation → Deployment
• ✅ Imbalanced Dataset Handling — Techniques for rare fraud class detection
• ✅ Multiple ML Models — Logistic Regression, Random Forest, XGBoost compared
• ✅ Saved Pipeline — .pkl model ready for production deployment
• ✅ Streamlit App — Interactive real-time fraud prediction interface
• ✅ Comprehensive Evaluation — Accuracy, Precision, Recall, F1, AUC-ROC
• ✅ Visual Analysis — Confusion Matrix, ROC Curve, Feature Importance

---

📊 Model Performance

Model	Accuracy	AUC Score
Logistic Regression	Baseline	—
Random Forest	98.7%	—
XGBoost	—	0.99
Overall Detection	94%	—

🏆 Best Model: Random Forest with 98.7% accuracy & XGBoost with 0.99 AUC

---

🛠️ Tech Stack

Technology	Purpose
Python 3.9+	Core programming language
Pandas & NumPy	Data manipulation & numerical operations
Scikit-learn	ML models, preprocessing & evaluation
XGBoost	Gradient boosting model
Matplotlib & Seaborn	Data visualization
Streamlit	Interactive web app deployment
Pickle (.pkl)	Model serialization & saving
Jupyter Notebook	Analysis & experimentation

---

📁 Project Structure

fraud-detection-ml-project/
│
├── analyse_model.ipynb              # Main analysis & training notebook
├── fraud_detection.py               # Streamlit app for real-time prediction
├── fraud_detection_pipline.pkl      # Saved ML pipeline (production ready)
├── Streamlit output 1.pdf           # Streamlit app screenshot — normal transaction
├── Streamlit output 2.pdf           # Streamlit app screenshot — fraud detected
├── .gitignore                       # Git ignore file
├── LICENSE                          # MIT License
└── README.md                        # Project documentation

---

🔬 How It Works

Raw Transaction Data
        ↓
Data Preprocessing
  → Handle missing values
  → Feature engineering
  → Handle class imbalance
        ↓
Model Training & Comparison
  → Logistic Regression (baseline)
  → Random Forest
  → XGBoost
        ↓
Model Evaluation
  → Accuracy, Precision, Recall, F1
  → ROC-AUC Curve
  → Confusion Matrix
        ↓
Save Best Model → fraud_detection_pipline.pkl
        ↓
Streamlit App → Real-time Prediction

---

🚀 How to Run

Step 1 — Clone Repository

git clone https://github.com/rakesh4407/fraud-detection-ml-project.git
cd fraud-detection-ml-project

Step 2 — Install Dependencies

pip install pandas numpy scikit-learn xgboost matplotlib seaborn streamlit jupyter

Step 3 — Run Analysis Notebook

jupyter notebook analyse_model.ipynb

Step 4 — Launch Streamlit App

streamlit run fraud_detection.py

---

📸 App Screenshots

Normal Transaction	Fraud Detected
See Streamlit output 1.pdf	See Streamlit output 2.pdf

---

💡 Key Learnings

• 🔄 Handling class imbalance is critical in fraud detection
• 🌲 Random Forest outperforms Logistic Regression for complex fraud patterns
• 📈 AUC-ROC is a better metric than accuracy for imbalanced datasets
• ⚡ XGBoost achieves near-perfect AUC (0.99) for fraud classification
• 🚀 Streamlit enables rapid ML model deployment without web dev knowledge

---

👨‍💻 Author

Rakesh G

BCA (H) — Artificial Intelligence & Data Science
K.R. Mangalam University, New Delhi | CGPA: 9.22/10
Dean's Award Recipient | IBM Certified Data Scientist

---

🏷️ Topics

python machine-learning fraud-detection scikit-learn xgboost random-forest streamlit data-science classification imbalanced-dataset fintech jupyter-notebook

---

📜 License

This project is licensed under the MIT License — see the LICENSE file for details.

---

⭐ If you found this useful, please star this repository!