AI-Driven and Explainable Detection of Fraud and Cyber-Related Risks in Critical Infrastructure Payment Networks

Abstract

Digital payment networks which face significant security breaches from financial fraud and cyber attacks have become essential for critical infrastructure sectors that encompass energy and healthcare and transportation and government systems. The detection of complex and evolving attack patterns which use coordinated methods fails to be achieved by traditional rule-based systems when they operate in environments with highly imbalanced transaction volumes. The research presents an explainable artificial intelligence framework which detects fraudulent and anomalous activities in payment systems that serve critical infrastructure. The framework uses real-world transactional data from the IEEE-CIS Fraud Detection Dataset to combine supervised and unsupervised learning techniques which include Logistic Regression and Random Forest and XGBoost for fraud classification and anomaly detection for irregular transaction behavior. To handle class imbalance the Synthetic Minority Oversampling Technique (SMOTE) is used whereas threshold optimization enhances the system's ability to identify different types of detection with better precision.

The framework uses SHAP to provide explainable AI which produces interpretable results that show how models make predictions and which features affect their predictions. The proposed approach successfully identifies high-risk transactions because it improves key evaluation metrics which include precision and recall and F1-score and AUC. The proposed framework provides a scalable and adaptable solution for securing payment systems that underpin critical infrastructure operations. This work creates financial fraud detection and cyber risk awareness systems which build resilient and transparent and intelligent security solutions that protect national economic stability and infrastructure systems