Credit Card Fraud Detection Using Autoencoders and Smote techniques based in Deep Neural Network (DNN)

The digital payment platforms have enhanced the ease of conducting financial transactions, but they have also led to a rise in fraudulent activities. The difficulty of detecting credit card fraud is exacerbated by the very skewed nature of the data – with a far greater number of legitimate transactions than fraudulent ones. This imbalance makes it harder to use traditional classification methods and necessitates the use of specialized techniques for identification of frauds with confidence. In this research, an end-to-end fraud detection system is explored, which considers several machine learning and deep learning solutions to detect suspicious transactions. Logistic Regression, Decision Tree, Random Forest, Support Vector Machine, XGBoost and LightGBM are the machine learning models considered, and 1D Convolutional Neural Network, Long Short-Term Memory network and Feed Forward Neural Network are the deep learning models explored. The proposed framework uses the data preprocessing techniques, such as feature normalization and class balancing using the Synthetic Minority Oversampling Technique (SMOTE) to enhance the recognition of the minority class. Furthermore, a hybrid approach called SentraFraud is presented that integrates isolation forest anomaly detection and LightGBM classification for improved fraud prediction. The developed system is made available as a web application using Flask and users can now analyze transaction details as well as receive real time fraud predictions. This suggested framework is designed to support an efficient, scalable and intelligent model to detect fraudulent payment activities by combining anomaly detection and advanced payment activity prediction models.