Detecting suspicious transactions on bitcoin networkusing unsupervised learning

Abstract

This research is the study and development of unsupervised learning algorithms to detect suspicious entities on the Bitcoin network. The objective is to develop a practical model for detecting anomalies in the Bitcoin network. This study was divided into two tasks, which are transaction and wallet address. The statistical techniques are applied for feature engineering and a Histogram-based Outlier Score (HBOS) and Isolation Forest (IForest) algorithms are trained and evaluated. The evaluations utilized were visualization, dual, and known-thieves evaluations. The result showed a similar detection for both algorithms. While HBOS has a higher wallet visualization score at 0.423, Isolation Forest yields better scores on transaction visualization, dual, and known-thieves evaluations with scores of 0.713, 0.681, and 0.035, respectively.

-