Anomalous Node Detection in Blockchain Networks Based on Graph Neural Networks.

Area of Science:

• Computer Science
• Cybersecurity
• Data Mining

Background:

• Blockchain technology's growth has led to increased fraudulent activities, threatening user assets.
• Blockchain transaction networks can be modeled as graphs, with fraudulent nodes being anomalous.
• Class imbalance in graph data, where anomalous nodes are a minority, challenges traditional detection methods.

Purpose of the Study:

• To address the class imbalance problem in detecting anomalous nodes in blockchain transaction networks.
• To propose a novel graph neural network method that improves upon existing graph data mining techniques for fraud detection.

Main Methods:

• Developed a new graph neural network method, SGAT-BC, enhancing the Graph Attention Network (GAT).
• Incorporated a subtree attention mechanism within the GAT framework.
• Integrated ensemble learning techniques: Bootstrap Aggregating (Bagging) and Categorical Boosting (CAT).

Main Results:

• The proposed SGAT-BC method demonstrated superior performance compared to existing baseline models.
• Experiments were conducted on four real-world blockchain transaction datasets, validating the method's effectiveness.
• SGAT-BC successfully overcomes the challenges posed by class imbalance in fraud detection.

Conclusions:

• The SGAT-BC method offers a robust solution for detecting fraudulent activities in blockchain networks.
• Combining GAT with subtree attention and ensemble learning is a promising approach for imbalanced graph data.
• This research contributes to enhancing the security of blockchain transactions and protecting user assets.