Privacy-preserving hybrid AI framework for secure fraud detection in .NET applications

Abstract

The fast growth of digital payment systems has made the problem of credit card fraud a major threat that demands a smart and secure system for detecting fraud. Certain cases of machine learning are known to be very bad at identifying fraud with severe imbalance in classes and complex behavior patterns in financial data. This paper suggests a Privacy-Aware Hybrid Ensemble Model (PAHEM) on a secure.NET web application framework for real-time fraud analytics. The proposed strategy involves the use of the Random Forest-based feature ranking, Extreme Gradient Boosting (XGBoost), and a Deep Neural Network (DNN) to identify both the linear and nonlinear trends in the transaction data. In order to eliminate the imbalance in classes, the Synthetic Minority Oversampling Technique (SMOTE) is used when training a model. The trained hybrid model is deployed as an encrypted analytics service that may be attached to the ASP.NET-based financial systems through the resources of the REST API, which will help to identify the fraud risk in real-time when processing transactions. Experimental analysis (on publicly available credit card fraud data) shows that the proposed framework can achieve a high detection accuracy of 99.94, 0.8229 precision, 0.8061 recall, 0.8144 F1-score, 0.9997 specificity, and ROC-AUC of 0.9819. These findings suggest that the offered PAHEM model suggests a high-quality detection of fraud and a small number of false positives. The combination of hybrid AI models with secure.NET web applications provides a scalable and privacy-conscious architecture that can be implemented in real-life financial transaction monitoring systems.