Enhancing Cryptocurrency Transparency: A Graph Neural Network Approach for Bitcoin Address Classification

Abstract

Cryptocurrencies, notably Bitcoin, have catalyzed a significant shift in digital financial systems. The inherent pseudonymity of blockchain complicates efforts towards transparency and security, presenting a crucial problem that this thesis aims to resolve by enhancing address classification. The relevance of this problem lies in the increasing necessity for compliance with global financial regulations and ensuring the integrity of transactions. Addressing this challenge involves overcoming significant difficulties such as the complexity of analyzing vast amounts of transaction data, the need for accurate data preprocessing, and the application of advanced machine learning techniques on non-traditional data structures like graphs. This research utilizes Graph Attention Networks (GATs) to classify Bitcoin addresses, a method chosen for its robustness in handling relational data and its capacity to focus selectively on the most informative parts of the transaction graph. The efficacy of this approach is demonstrated through controlled experiments, where the GATs achieved an accuracy of 92.87%, a precision of 89.35%, a recall of 92.87%, and an F1 score of 90.17%. These results significantly improve upon previous internal benchmarks and confirm the model’s capability to enhance transparency in Bitcoin transactions. Furthermore, this work contributes a novel open-source Extract, Transform, Load (ETL) process tailored for blockchain data, fostering improved analytical transparency, and aiding regulatory and forensic analysis. The findings propose practical applications in financial technology, moving beyond theoretical discourse into actionable insights.