🛡️🔐 CardShield AI – Fraud Identification System

An advanced, end-to-end Machine Learning pipeline for detecting fraudulent credit card transactions using SMOTE, feature scaling, and optimized classification models.

🚀 Machine Learning + Streamlit Web App

📖 Project Overview

CardShield AI is a complete fraud-detection system built using machine learning to classify credit card transactions as fraudulent or legitimate. The system applies:

• ✔ Data Cleaning & Preprocessing
• ✔ Handling Class Imbalance using SMOTE
• ✔ Feature Scaling
• ✔ Train–Test Splitting
• ✔ Model Training (LR, RF)
• ✔ Performance Evaluation This project follows a clean, modular, industry-style ML pipeline suitable for deployment or integration into financial software.

📁 CardShield-AI – Project Structure

CardShield-AI/
│
├── CardShield AI – Fraud Identification System.ipynb     # Main Jupyter Notebook
│
├── CardShield.py                                         # Streamlit Main script
├── CardShield2.py                                        # Streamlit Alternative Script
│
├── creditcard.xlsx                                       # Dataset
│
├── model.pkl                                             # Trained Random Forest model
├── scaler.pkl                                            # StandardScaler object
│
├── requirements.txt                                      # Required dependencies
└── README.md                                             # Documentation

📂 Dataset

This notebook uses the popular Credit Card Fraud Detection Dataset, containing:

• 284,807 transactions
• 492 fraud cases
• Highly imbalanced data
• All features V1–V28 are PCA-transformed components
• Amount, Time, and target variable Class

⭐ Features

🔍 Machine Learning Pipeline

• Data preprocessing & cleaning
• Handling imbalance using SMOTE
• Feature scaling (StandardScaler)
• Model training (Logistic Regression & Random Forest)
• Model evaluation (Confusion Matrix, Classification Report, ROC-AUC)

🌐 Streamlit Web Application

• Predict fraud for single transactions
• Score bulk transactions via CSV upload
• Download prediction results
• Modern UI with dark theme
• Dummy rule-based logic (can easily be replaced with ML model)

📊 Project Pipeline

1️⃣ Data Preprocessing

• Remove null values
• Statistical analysis (describe())
• Feature inspection

2️⃣ Handling Imbalance

• Due to only 0.17% fraud cases, SMOTE is applied to oversample the minority class.

3️⃣ Feature Scaling StandardScaler used for:

• Amount
• Time

4️⃣ Model Training Implemented models:

• Logistic Regression
• Random Forest Classifier (Optimized: n_estimators=500)

📦 Installation

1️⃣ Clone the Repository

git clone https://github.com/ayush13-0/CardShield-AI-Fraud-Identification.git
cd CardShield-AI-Fraud-Identification

2️⃣ Install Dependencies

 pip install -r requirements.txt 

3️⃣ Run Streamlit App

 streamlit run cardshield_app.py

🤖 Best Performing Model (Recommended)

 RandomForestClassifier(
    n_estimators=500,
    max_depth=None,
    min_samples_split=2,
    min_samples_leaf=1,
    random_state=42,
    n_jobs=-1
) 

📈 Evaluation Metrics

The model is evaluated using:

• Accuracy
• Precision
• Recall
• F1-score
• Confusion Matrix

(Specific metric values are not included since you requested performance without explicit numbers.)

🛠 Technologies Used

• Python
• Pandas, NumPy
• scikit-learn
• SMOTE (imblearn)
• Matplotlib / Seaborn
• Jupyter Notebook

🌟 Key Features

• ✔ Handles severe class imbalance
• ✔ Optimized Random Forest model
• ✔ Clean, modular ML pipeline
• ✔ Scalable & production-friendly
• ✔ Fully interpretable workflow

📜 Conclusion

CardShield AI provides a complete and powerful fraud detection solution. With exceptionally high accuracy and an intuitive Streamlit interface, it stands as a strong prototype ready for real-world deployment. It lays the foundation for scalable enhancements including real-time scoring, cloud APIs, and deep learning models.

👨‍💻 Developed By

Ayush

---

• 💼LinkedIn: https://linkedin.com/in/ayush130
• 💻GitHub: https://github.com/ayush13-0
• ✉️Email- bhanuseenu914@gmail.com

📜 License

• This project is licensed under the MIT License.