💳 Financial Fraud Detection using Machine Learning

📖 Project Overview

This project builds a high-performance fraud detection system using advanced feature engineering, SMOTE balancing, and machine learning models to identify fraudulent financial transactions.

The goal is to accurately classify transactions as fraud or non-fraud while minimizing financial risk.

---

🎯 Problem Statement

To detect fraudulent transactions by analyzing:

• Transaction type
• Transaction amount
• Account balance changes
• Engineered balance-difference features

---

🛠️ Technologies Used

• Python
• Pandas
• NumPy
• Scikit-learn
• XGBoost
• SMOTE (Imbalanced Learning)
• Matplotlib & Seaborn
• SHAP (Model Explainability)

---

⚙️ Methodology

1. Exploratory Data Analysis (EDA)
2. Feature Engineering
   • org_diff
   • dest_diff
   • amount_log
   • zero balance indicators
   • one-hot encoding
3. Handling Class Imbalance using SMOTE
4. Train-Test Split (Stratified 80/20)
5. Model Training:
   • Logistic Regression
   • Random Forest
   • XGBoost
6. Model Evaluation:
   • Precision
   • Recall
   • F1-Score
   • ROC-AUC
   • PR-AUC
7. SHAP Explainability

---

📊 Model Performance

🏆 Best Model: XGBoost

• Precision: 99.13%
• Recall: 99.56%
• F1-Score: 99.34%
• ROC-AUC: 99.99%
• PR-AUC: 99.95%

XGBoost achieved the highest recall and PR-AUC, making it the most effective fraud detection model.

---

🧠 Key Fraud Indicators

• Origin balance difference (org_diff)
• Log-transformed transaction amount
• Zero-balance behavior
• CASH_OUT & TRANSFER transaction types
• Abnormal balance drops

---

💼 Business Impact

• Reduces financial losses
• Improves fraud detection recall
• Enables real-time risk scoring
• Supports scalable fraud prevention systems

---

---

🚀 How to Run

1. Clone repository
2. Install dependencies
3. Run Jupyter notebook

---

👤 Author

Subodh Kumar
Machine Learning & Data Science Enthusiast