AI GRX model for real-time risk assessment and forecasting in blockchain ecosystems

DOI: 10.31673/2412-9070.2026.024502

Abstract

The article presents the development of a novel GRX model for real-time risk assessment and forecasting in blockchain ecosystems. The model leverages a combination of Graph Neural Networks (GNNs), temporal deep learning architectures, and explainability mechanisms. The increasing structural complexity of cryptocurrency transaction networks, the emergence of new schemes for evading financial oversight, and the growing number of highly dynamic relationships between network participants significantly reduce the effectiveness of traditional rule-based AML approaches. The proposed GRX model captures both local transaction patterns and global structures of the blockchain graph, enabling adaptive risk assessment for addresses, clusters, and transactions.
The methodological foundation is a hybrid architecture that integrates classical GNN operators (GCN, GAT, GraphSAGE) with a temporal LSTM-based component, allowing the model to track the evolution of risk over time and predict future risk values. The system supports the analysis of complex dynamic AML-graphs, including branching, cyclic behavior, aggregation nodes, and latent money laundering typologies. Furthermore, the architecture includes an explainability (XAI) module that incurporates GAT attention mechanisms and graph-based interpretability methods to quantify the contribution of specific edges, nodes, and structural features to the final risk score.
The paper presents the mathematical formulation of the GRX model, describes the algorithmic architecture, introduces a prototype for real-time transaction-stream processing, and provides visualizations of high-risk graph structures. Experimental results demonstrate improved detection accuracy of risky addresses and transactions, a reduction in false positives, and significantly enhanced predictive performance compared to traditional methods. The proposed GRX architecture is flexible, scalable, and suitable for integration into modern AML/KYT platforms used by financial institutions, exchanges, payment providers, and blockchain analytics services.

Keywords: blockchain, artificial intelligence, neural networks, crypto-assets, graph neural networks (GNN), dynamic AML graphs, risk scoring, transaction analysis, real-time risk assessment, AML/KYT systems, blockchain analytics.