https://dl.acm.org/doi/epdf/10.1145/3511808.3557598 cyber fraud

Unique Title: “InterpGraph: Search-Based Explainable Threat Intelligence in Financial Cyberfraud Networks” Abstract: Financial cyberfraud remains an escalating threat in online ecosystems, requiring detection models that are not only accurate but also interpretable for security auditing and compliance. Current graph-based fraud detection approaches suffer from limited transparency and rely heavily on manually defined relational patterns, hindering their deployment in dynamic, adversarial environments. This paper introduces InterpGraph, a novel detection framework that leverages neural architecture search (NAS) over heterogeneous transaction graphs to uncover latent, malicious interaction patterns without human supervision. The system automates the discovery of meta-graph schemas that encode adversarial behavior pathways, enabling both high-precision fraud identification and intrinsic model interpretability. Evaluated on real-world e-commerce fraud datasets, InterpGraph outperforms state-of-the-art baselines, achieving up to 92.28% F1-macro and 97.36% AUC, while providing semantic explanations that align with known attack strategies such as click farming and review spam. The framework ensures real-time adaptability in fraud detection pipelines, making it a deployable solution for continuous threat monitoring in financial cybersecurity operations.