Credit Card Fraud Detection using Deep Neural Networks with Autoencoders and SMOTE Techniques

Abstract

This paper presents a hybrid credit card fraud detection system that integrates both machine learning and deep learning techniques to accurately identify fraudulent transactions. Due to the highly imbalanced nature of financial datasets, the Synthetic Minority Over-sampling Technique (SMOTE) is applied to balance the distribution between legitimate and fraudulent transactions. The system utilizes multiple models, including Random Forest and XGBoost for classification, along with deep learning architectures such as Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), and Feedforward Neural Networks (FNN) to capture complex patterns in transaction data. Additionally, an autoencoder-based anomaly detection mechanism is incorporated to identify previously unseen fraudulent behaviors. A Flask-based web application is developed to provide a user-friendly interface for dataset upload, model selection, and real-time prediction. The system evaluates performance using standard metrics such as accuracy, precision, recall, and F1-score. Experimental results demonstrate that the proposed hybrid approach significantly improves fraud detection performance, reduces false negatives, and enhances the overall reliability of the system. This work provides a scalable and efficient solution for real-time fraud detection in financial applications.