Skip to content
Conference

XGLF-BML: An Explainable Graph Learning Framework with Adaptive Attention for Blockchain Money Laundering Detection using Heterogeneous Transaction Graphs

Aug 2026 · 2026 International Conference on Secure Information Systems and Technologies (ICSIST) · pp. 698-703 · 0 citations · 14 references

Abstract

Cryptocurrency is making it much easier for crooks to launder money. These schemes rely on the anonymity and distributed nature of blockchains. Currently, strategies to tackle this issue usually depend on regulatory rules or machine learning models which have their flaws. They might not cover all crypto networks and can be complex to figure out. This paper, however, presents XGLF-BML. It’s crafted to detect money laundering using Graph Attention Networks version 2 (GATv2) and GraphSAGE inductive learning. Plus, it incorporates a unique loss function and an explainability module consisting of GNNExplainer and SHAP. The process starts with analyzing plain blockchain data, applying intelligent processing through attention aggregation, and wrapping up with human-readable outputs. Tests run on the Elliptic Bitcoin Dataset and another anti-money laundering dataset showed promising results. The precision hit 0.961, recall reached 0.954, F1-score came in at 0.957, and AUC-ROC peaked at 0.978. Compared to the best previous approaches, XGLF-BML delivered a notable upgrade ranging from 3.2 to 10.4 percentage points. Perhaps the greatest advantage is its ability to work in real time and provide understandable reports for authorities. This aligns perfectly with what FinCEN and FATF require.

View source

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.