Credit Card Fraud Detection

1. Project Overview Our project focuses on Credit Card Fraud Detection, leveraging machine learning to identify real-time fraudulent transactions. Fraudulent transactions cost businesses and consumers billions annually, and our goal is to build a highly accurate, scalable, and interpretable fraud detection system that minimizes false positives while catching actual fraud. Target Users: 🔹 Banks & financial institutions 🏦 🔹 Payment processors 💳 🔹 E-commerce platforms 🛒 Key Features & Problems Solved: ✅ Detects fraud in real time to reduce financial losses ✅ Reduces false positives to avoid blocking legitimate transactions ✅ Provides explainability to increase trust in fraud detection models 2. How We Built It 🔹 Dataset: Used the Kaggle Credit Card Fraud Detection dataset (imbalanced real-world transactions). 🔹 Preprocessing & Feature Engineering: Removed duplicates, handled class imbalance using BorderlineSMOTE (0.3) Engineered new features like cyclical time representation (Hour_sin, Hour_cos) Scaled features with RobustScaler & applied Winsorization for outlier handling Applied Tomek Links to refine the dataset 🔹 Modeling: Trained multiple models such as Balanced Random Forest, Logistic Regression, XGBoost, and autoencoder Evaluated models using Precision, Recall, F1-score, and AUC-ROC 🔹 Deployment: API built using FastAPI Model deployed on a cloud server for real-time fraud detection 3. Next Steps 🔹 Optimize Model Performance: Further improve feature engineering and hyperparameter tuning 🔹 Scalability: Explore real-world deployment on GCP for handling high transaction volumes 🔹 Business Integration: Test the model with simulated live transactions and enhance API responses We’re excited about the potential impact of this project and look forward to expanding its capabilities! 🚀