CreditCardGuard

Detecting credit card fraud using machine learning and real financial data

Project Overview

CreditCardGuard is a machine learning project that leverages real-world credit card transaction data to detect fraudulent activities. Using advanced classification algorithms and data preprocessing techniques, the system achieves high accuracy in identifying suspicious transactions while minimizing false positives.

Dataset Information

Transactions

284,807 records

Features

30 PCA-transformed + Amount + Class

Class Distribution

99.83% Normal, 0.17% Fraud

Data Preprocessing

Removed duplicate transactions to ensure data quality
Standardized the "Amount" column using StandardScaler
Dropped the "Time" column as it wasn't relevant for prediction
Applied both undersampling and SMOTE to handle severe class imbalance

Model Performance

Model	Accuracy	Precision	Recall	F1 Score
Logistic Regression	0.999	0.887	0.604	0.719
Decision Tree	0.999	0.686	0.769	0.725

Model Performance Visualizations

Class Distribution

Confusion Matrices

ROC Curves

F1 Score Comparison

Key Takeaways

Class imbalance significantly impacts model performance - proper handling is crucial
Logistic Regression shows high precision but lower recall, suitable when false positives are costly
Decision Trees offer better recall, catching more fraud cases but with more false alarms
Feature engineering and selection play a crucial role in model performance

Sample Inference Code

import joblib

# Load the trained model
model = joblib.load('creditcard_model.joblib')

# Prepare transaction data (after preprocessing)
transaction = preprocess_transaction(raw_data)

# Get prediction
prediction = model.predict(transaction)
probability = model.predict_proba(transaction)

if prediction[0] == 1:
    print(f"Fraud detected! Confidence: {probability[0][1]:.2f}")

Future Improvements

Real-time Alerts

Implement a real-time transaction monitoring system with immediate fraud alerts

Web Interface

Develop a Flask/Streamlit frontend for easy model interaction

Notification System

Add email/SMS alerts for suspicious transactions