Big Data Analytics for Anomaly Detection in Blockchain

The ability to make fast and automatic transactions anonymously without intermediary institutions offered by blockchain technology has made this technology widespread in the use of money transfer. With its widespread use, it has also become a preferred transfer method for illegal transactions such as tax evasion and the sale of illegal goods. This has increased the importance of anomalous transactions and anomalous user detection studies in transactions. In this study, a solution is presented that enables the detection of anomalies and anomalous users in blockchain transfers by using big data storage and data processing tools. A scalable structure has been created in order to deal with the increase in the size and frequency of blockchain transfers with the use of big data technologies. In the solution, first of all, the users making the transfers and the transfer transactions are clustered using the K-Means method. Transfer information and attributes that represent the blockchain network as a graph are used as features. After clustering, anomalies were found by measuring the distance of the elements from the cluster center. The performance of the clustering was calculated using the silhouette score. When the results are analyzed, the model that reduces the size to two with principal component analysis for clustering users showed the highest scores with a silhouette score of 0.51, and the model that reduces the size of graph-based attributes with principal component analysis for clustering transfers showed the highest scores with a silhouette score of 0.87. These models were applied to real time transactions.