Adversarially Robust and Explainable Graph Neural Networks for Fraud Detection in Dynamic Blockchain Networks: A Problem-Based Systematic Literature Review

Abstract

Graph Neural Networks (GNNs) have emerged as an effective approach for blockchain fraud detection because they can model complex relationships among transactions and participating entities. However, existing GNN-based approaches remain limited by vulnerability to adversarial manipulation, limited explainability, continuously evolving transaction graphs, benchmark-data constraints, and the absence of standardized evaluation practices. This study presents a Problem-Based Systematic Literature Review (PBSLR) of adversarially robust and explainable GNNs for fraud detection in dynamic blockchain networks. The review covers studies published between 2021 and 2026 and searches eight academic databases: IEEE Xplore, ACM Digital Library, ScienceDirect, SpringerLink, Wiley Online Library, Scopus, Web of Science, and Google Scholar. A total of 412 records were initially identified, 77 full-text studies were assessed for eligibility, and 73 studies were included in the final systematic synthesis. The review analyses the literature across blockchain fraud characteristics, GNN-based detection, adversarial threats, explainability, dynamic graph learning, benchmark datasets, evaluation practices, scalability, and deployment challenges. The synthesis shows that existing research remains fragmented, with limited integration of adversarial robustness, explainability, and dynamic graph learning within a unified fraud detection framework. The review further identifies persistent limitations involving dynamic and labelled benchmark datasets, adversarial evaluation, cross-platform generalisation, standardized robustness and explainability metrics, scalability, and reproducibility. Based on these findings, a research roadmap is proposed to guide the development of robust, interpretable, adaptive, scalable, and deployable GNN-based fraud detection systems for dynamic blockchain environments.