---

💳🚨 Credit Card Fraud Detection System Project

📌 Project Overview

Credit card fraud detection is a highly imbalanced classification problem where fraudulent transactions represent a very small fraction of all transactions, yet carry significant financial and reputational risk. This project presents an end-to-end machine learning pipeline for detecting fraudulent credit card transactions, with a strong emphasis on model performance, threshold optimization, and interpretability.

The goal is not only to build an accurate fraud detection model, but also to ensure that predictions are transparent, explainable, and aligned with real-world decision-making.

---

🎯 Objectives

• Detect fraudulent transactions with high recall while maintaining acceptable precision
• Address severe class imbalance using appropriate modeling strategies
• Select decision thresholds based on business-aligned trade-offs
• Provide global and local explanations of model predictions using SHAP
• Produce a reproducible, portfolio-grade machine learning workflow

---

📂 Dataset

• Source: Kaggle Dataset Credit Card Fraud Detection

• Description: Anonymized transaction features derived from PCA transformations, along with transaction amount and class label

• Target Variable:

  • 0 → Legitimate transaction
  • 1 → Fraudulent transaction

• Key Challenge: Extreme class imbalance (~0.17% fraud cases)

---

🛠️ Methodology

1️⃣ Data Preprocessing

• Train–test split with stratification
• Feature scaling where appropriate
• Careful handling of class imbalance

2️⃣ Modeling Approaches

• Random Forest Classifier

  • Baseline model
  • Explored class weighting and threshold adjustment

• LightGBM Classifier (Final Model)

  • Gradient-boosted decision trees
  • Optimized for imbalanced classification
  • Superior recall–precision trade-off

3️⃣ Model Evaluation

Models were evaluated using metrics suitable for imbalanced data:

• Precision
• Recall
• F1-score
• ROC-AUC
• Precision–Recall curves

Decision thresholds were selected based on:

• Precision-constrained recall
• F1-score maximization
• Business cost considerations

---

🧠 Model Interpretability (SHAP)

To ensure transparency and trust, SHAP (SHapley Additive exPlanations) was used to interpret model predictions.

Global Interpretability

• Feature importance analysis to identify key fraud indicators
• SHAP summary plots to understand direction and magnitude of feature effects

Local Interpretability

• Transaction-level explanations for individual fraud cases
• Clear visualization of features pushing predictions toward fraud or legitimacy

All explanations focus on the fraud class (class 1) and are aligned with the final decision threshold.

---

🏆 Results Summary

• LightGBM achieved strong recall on fraudulent transactions while maintaining reasonable precision
• SHAP analysis confirmed that the model relies on a consistent and interpretable set of fraud indicators
• The final system balances performance, explainability, and reproducibility

---

📁 Project Structure

├── dataset/
│   └── archive.zip
├── notebooks/
│   └── Credit_Card_Fraud_Detection_System.ipynb
├── models/
│   ├── lgbm/
|   |     ├── lgbm_baseline.pkl
|   |     ├── lgbm_tuned.pkl
|   |     └── lgbm_metadata.json
│   └── random_forest/
|          ├── rf_metadata.json
|          └── rf_model.pkl 
├── README.md
└── requirements.txt

---

🚀 How to Run the Project

1. Clone the repository:

git clone https://github.com/SaintJeane/credit_card_fraud_detection_system_for_credit_card_transactions.git

2. Install dependencies:

pip install -r requirements.txt

3. Open the notebook:

jupyter notebook notebooks/Credit_Card_Fraud_Detection_System.ipynb

---

📦 Tools & Technologies

• Python
• scikit-learn
• LightGBM
• SHAP
• NumPy
• pandas
• matplotlib / seaborn
• joblib

---

📈 Key Takeaways

• Imbalanced classification requires careful metric selection, not accuracy alone
• Decision thresholds are as important as the model itself
• Interpretability is essential for deploying fraud detection systems in practice
• Combining performance with explainability leads to more trustworthy models

---

This project was developed as a portfolio-grade demonstration of applied machine learning for fraud detection, with a focus on real-world constraints and interpretability.

---