Modeling the process of address clustering in a blockchain based on Bayesian optimization of heuristic weights

This article examines the problem of blockchain address clustering, a key method in cryptocurrency forensics used in investigating digital asset crimes. An approach combining three clustering heuristics is proposed: Multi-Input Heuristic (MIH), Change Address Heuristic (CAH), and Temporal Pattern Heuristic (TPH). Weighted voting is used for the combination, and Bayesian optimization with a Gaussian process as a surrogate model and an Expected Improvement function is applied to select the weights and the merging threshold. A mathematical clustering model is formalized using AT-graphs, A-graphs, and E-graphs, and the stages of transition from a transaction graph to an entity graph are described. Three experiments are conducted on real data from the maltemoeser dataset. In the first experiment (500 transactions, 1562 addresses), combining the heuristics allowed us to achieve an F1-measure (harmonic mean between precision and recall) of 0.9284, which corresponds to the level of the best individual heuristic (CAH: 0.9284). In the second experiment (775 transactions, 2850 addresses), F1 = 0.8251 was obtained, which exceeds the performance of individual heuristics (MIH: 0.8151, CAH: 0.8184, TPH: 0.0838). In the third experiment (990 transactions, 3445 addresses), F1 = 0.8146 was obtained, which also exceeds the performance of individual heuristics. The experimental results support the presence of a synergistic effect of the proposed approach. In some cases, combining the heuristics makes it possible to partially compensate for the shortcomings of individual heuristics and achieve higher clustering quality. Changes in the data structure with an increase in the sample size lead to a redistribution of the heuristic weights, which reflects the adaptability of the proposed method. The method's practical applicability to real-world investigations is substantiated: reference labeling is required only during the calibration stage, after which the weights are fixed for application to new data. A comparative analysis with alternative machine learning approaches is conducted.

1. Tisen O.N. The mechanism of committing crimes involving the use of cryptocurrencies: based on the results of an analysis of the investigative and judicial practice. Russian Investigator. 2024;(5):7–11. (In Russ.). https://doi.org/10.18572/1812-3783-2024-5-7-11

2. Harrigan M., Fretter Ch. The Unreasonable Effectiveness of Address Clustering. In: 2016 Intl IEEE Conferences on Ubiquitous Intelligence & Computing, Advanced and Trusted Computing, Scalable Computing and Communications, Cloud and Big Data Computing, Internet of People, and Smart World Congress, 18–21 July 2016, Toulouse, France. IEEE; 2016. P. 368–373. https://doi.org/10.1109/UIC-ATC-ScalCom-CBDCom-IoP-SmartWorld.2016.0071

3. Möser M., Narayanan A. Resurrecting Address Clustering in Bitcoin. In: Financial Cryptography and Data Security: 26th International Conference (FC 2022), 02–06 May 2022, Grenada. Cham: Springer; 2022. P. 386–403. https://doi.org/10.1007/978-3-031-18283-9_19

4. Caringella M., Violante F., De Lucci F., Galantucci S., Costantini M. BACH: A Tool for Analyzing Blockchain Transactions Using Address Clustering Heuristics. Information. 2024;15(10):589. https://doi.org/10.3390/info15100589

5. Liu F., Li Zh., Jia K., et al. Bitcoin Address Clustering Based on Change Address Improvement. IEEE Transactions on Computational Social Systems. 2024;11(6):8094–8105. https://doi.org/10.1109/TCSS.2023.3239031

6. Rasmussen C.E., Williams Ch.K.I. Gaussian Processes for Machine Learning. Cambridge: MIT Press; 2005. 272 p. https://doi.org/10.7551/mitpress/3206.001.0001

7. Shahriari B., Swersky K., Wang Z., Adams R.P., de Freitas N. Taking the human out of the loop: a review of Bayesian optimization. Proceedings of the IEEE. 2016;104(1):148–175. https://doi.org/10.1109/JPROC.2015.2494218

8. Kamuhanda D., Cui M., Tessone C.J. Illegal Community Detection in Bitcoin Transaction Networks. Entropy. 2023;25(7):1069. https://doi.org/10.3390/e25071069

9. Huang Zh., Huang Y., Qian P., Chen J., He Q. Demystifying Bitcoin address behavior via graph neural networks. In: 2023 IEEE 39th International Conference on Data Engineering (ICDE), 03–07 April 2023, Anaheim, CA, USA. IEEE; 2023. P. 1747–1760. https://doi.org/10.1109/ICDE55515.2023.00137

10. Schnoering H., Vazirgiannis M. Bitcoin research with a transaction graph dataset. Scientific Data. 2025;12(1):404. https://doi.org/10.1038/s41597-025-04684-8