Clustering Blockchain Wallet Behaviour Using K-Means and DBSCAN for Risk Profiling and Address Segmentation

Blockchain networks generate high-dimensional transactional data with diverse and irregular wallet behaviours. Understanding these behavioural patterns is essential for improving security monitoring, anomaly detection, and risk assessment within decentralized systems. This study applies two unsupervised machine learning algorithms, K-Means and DBSCAN, to analyse 303 blockchain wallet records using key attributes such as BlockHeight, UnixTimestamp, Block Density, Coin Day Weight, and Stake Distribution Rate. K-Means successfully identified three distinct behavioural clusters consisting of Cluster 1 with 200 wallets, Cluster 2 with 100 wallets, and Cluster 0 with 3 highly anomalous wallets. Numerical analysis revealed clear differences across clusters, including mean BlockHeight values of 5.5 million for Cluster 1, 15.4 million for Cluster 2, and 10.9 million for Cluster 0, along with Block Density percentages of 19.35, 48.90, and 60.00, respectively. DBSCAN further exposed behavioural complexity by detecting more than 90 noise points that represent irregular or outlier activity patterns and several small micro-clusters not captured by K-Means. PCA visualizations confirmed strong separation between clusters and highlighted the unique positioning of anomalous wallets. The combined use of centroid-based and density-based clustering provides a robust analytical foundation for profiling blockchain wallet behaviour, supporting more effective anomaly detection, risk classification, and address segmentation.