AI-ML Project: Credit Card Fraud Detection

Overview

This project implements a Credit Card Fraud Detection System using a machine learning approach. The system detects fraudulent transactions in a highly imbalanced dataset using a Random Forest Classifier with threshold tuning for optimal precision and recall.

The model is trained on the publicly available Credit Card Fraud dataset, which contains 284,807 transactions with 28 anonymized features (V1–V28), Time, Amount, and a target Class (0 = normal, 1 = fraud).

---

Project Structure

AI-ML-Project/
│
├─ backend/
│   ├─ ai_model/
│   │   ├─ ai-model.py           # Python script to train and evaluate the model
│   │   ├─ fraud_model.pkl       # Serialized trained model (~361 MB)
│   │   ├─ scale.pkl             # StandardScaler for preprocessing
│   │   └─ creditcard.csv        # Sample dataset for testing
│   ├─ controller/
│   │   └─ transaction.py        # Handles transaction processing
│   ├─ core/
│   │   └─ database.py           # Database connection utility
│   ├─ routes/
│   │   └─ transaction_routes.py # API endpoints for transactions
│   └─ main.py                   # FastAPI main entrypoint
│
├─ frontend/
│   ├─ src/
│   │   └─ app/                  # React components
│   ├─ node_modules/
│   └─ package.json
│
└─ .gitignore

---

Model Details

• Algorithm: Random Forest Classifier
• Features: 28 PCA components (V1–V28), Time, Amount
• Target: Class (0 = Normal, 1 = Fraud)
• Preprocessing: StandardScaler for normalization
• Imbalanced Handling: SMOTEENN for oversampling minority class
• Evaluation Metrics: Precision, Recall, F1-score

Confusion Matrix (Threshold 0.2):

[[56857    7]
 [   15    83]]

---

Setup & Installation

1. Clone the repository:

git clone https://github.com/SahilGupta721/AI-ML-Project.git
cd AI-ML-Project

2. Set up the Python virtual environment:

python -m venv virtual-env
source virtual-env/Scripts/activate   # Windows
# or
source virtual-env/bin/activate       # Mac/Linux

3. Install dependencies:

pip install -r requirements.txt

4. Run the model script for evaluation or training:

cd backend/ai_model
python ai-model.py

---

Usage

• Loading the model in Python:

import joblib

model = joblib.load('fraud_model.pkl')
scaler = joblib.load('scale.pkl')

• Predicting on new transaction data:

import pandas as pd

# Load new transaction data
new_data = pd.read_csv('new_transactions.csv')
new_scaled = scaler.transform(new_data)
predictions = model.predict(new_scaled)

• API Integration: Use the FastAPI backend to expose endpoints for predictions. The transaction_routes.py handles requests.

---

Notes

• The dataset is highly imbalanced; threshold tuning is recommended for precision/recall trade-off.
• Large files (fraud_model.pkl) may exceed GitHub file limits; consider using Git LFS.
• The project uses Python 3.10+, Scikit-Learn, Imbalanced-learn, and FastAPI.

---

License

This project is licensed under the MIT License.

---

References

• Credit Card Fraud Detection Dataset - Kaggle
• Scikit-learn Documentation
• Imbalanced-learn Documentation
• FastAPI Documentation