A Cybersecurity Evaluation Framework for Fraud Detection: Integrating STRIDE Threat Modelling, Explainable Alerts and Anomaly Gating

Abstract

Financial fraud continues to evolve in complexity, challenging traditional detection methods. Machine learning has provided powerful tools but it remains vulnerable to adversarial manipulation, requires transparency and may operate disconnected from established cybersecurity frameworks. This study proposes a hybrid evaluation framework which combines selective STRIDE threat analysis, SHAP-based explainable alerts and an anomaly-gating mechanism that leverages on Isolation Forest scores. The study uses IEEE-CIS dataset to uncover critical vulnerabilities in financial detection such as identity spoofing and feature tampering. The Model that was used in this study integrated explainable alerts to improve analyst decision-making and operational transparency. Despite severe recall trade-offs anomaly gating effectively reduces false positives and workload demonstrating the practical difficulty of balancing precision and resilience. The results of this study highlight that effective fraud detection requires moving beyond accuracy-focused models by integrating frameworks that embed explainability, threat modeling and cybersecurity principles. This work contributes a realistic blueprint for moving fraud detection research beyond narrow accuracy metrics toward integrated, security-aware frameworks that prioritize explainability, resilience and operational integration.

Keywords: financial fraud detection, STRIDE threat modelling, explainable AI, SHAP explanations, anomaly detection, Isolation Forest, adversarial robustness, cybersecurity resilience, Security Operations Center (SOC), SIEM integration

DOI: 10.7176/ISDE/15-06

Publication date: October 31^st 2025