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

Abstract

Blockchain technology has revolutionized digital financial systems through decentralized, transparent, and secure transactions. However, the increasing sophistication of blockchain systems has coincided with the emergence of complex illicit activities, including ransomware transactions, Ponzi schemes, phishing, and money laundering. Traditional fraud-detection techniques may be inadequate for blockchain transaction networks because these networks are complex, dynamic, and highly interconnected. Graph Neural Networks (GNNs) have emerged as a promising approach because of their ability to capture relationships within graph structures and their evolving behavioural patterns. This paper presents a systematic literature review of adversarially robust and explainable fraud-detection solutions based on GNNs for dynamic blockchain networks. The review was conducted in accordance with the PRISMA guidelines and critically examined studies published between 2018 and 2026. The study provides an integrated synthesis of graph representation learning, dynamic graph learning, adversarial robustness, explainable artificial intelligence (XAI), trustworthy AI, benchmark datasets, evaluation strategies, and regulatory compliance within the context of blockchain fraud detection. The review indicates that recent advances in Graph Convolutional Networks, Graph Attention Networks, GraphSAGE, Graph Transformers, and Temporal Graph Neural Networks have demonstrated promising improvements in fraud-detection performance across various experimental settings. However, challenges remain in achieving scalability, robustness against adversarial attacks, explainability, standardized evaluation, and real-time deployment. The review identifies major research gaps in blockchain fraud detection and examines the strengths, limitations, applicability, and maturity of existing techniques. Finally, based on the synthesized findings, a research roadmap is proposed for developing blockchain fraud-detection systems with improved explainability, scalability, robustness, and trustworthiness using GNNs. The findings provide actionable insights and recommendations for researchers and practitioners developing blockchain security solutions.