Graph convolution network for fraud detection in bitcoin transactions.

Area of Science:

• Computer Science
• Cybersecurity
• Financial Forensics

Background:

• Anti-money laundering (AML) is challenged by the pseudonymous nature of cryptocurrency transactions.
• Criminals exploit cryptocurrency for illicit financial activities, necessitating advanced detection methods.
• Machine learning and deep learning offer promising approaches to identify anomalies in blockchain data.

Purpose of the Study:

• To evaluate the effectiveness of various machine learning algorithms for detecting illicit cryptocurrency transactions.
• To specifically assess the performance of a Graph Convolutional Network (GCN) on the Elliptic Bitcoin Dataset.
• To compare the GCN model's efficacy against existing AML detection models.

Main Methods:

• Utilized the Elliptic Bitcoin Dataset, a graph dataset of labeled and unlabeled blockchain transactions.
• Implemented and compared Logistic Regression, Long Short-Term Memory (LSTM), Support Vector Machine (SVM), Random Forest, and Graph Convolutional Network (GCN) models.
• Evaluated model performance using accuracy, Area Under the Receiver Operating Characteristic Curve (AUC), and Root Mean Squared Error (RMSE).

Main Results:

• The proposed GCN model achieved an accuracy of [Formula: see text], an AUC of 0.9444, and an RMSE of 0.1123.
• The GCN model demonstrated superior performance compared to other evaluated machine learning algorithms.
• Results indicate the GCN model is more effective than previously proposed models, including the one by Weber et al. (2019).

Conclusions:

• Graph Convolutional Networks (GCNs) are highly effective for detecting illicit cryptocurrency transactions.
• The developed GCN model provides a robust solution for financial forensics in the context of anti-money laundering.
• This research highlights the potential of graph-based deep learning for securing blockchain ecosystems.