Bitcoin abnormal transaction detection model based on machine learning

This article is devoted to the development of a reliable model for detecting abnormal bitcoin transactions that may be involved in money laundering and illegal trafficking of goods and services. The article proposed a model for detecting abnormal bitcoin transactions based on machine learning. For training and evaluation of the model, the Elliptic dataset is used, consisting of 46564 Bitcoin transactions: 4545 of "illegal" and 42019 of "legal" . The proposed model for detecting abnormal bitcoin transactions is based on various machine learning algorithms with the selection of hyperparameters. To evaluate the proposed model, we used the metric of accuracy, precision, recall, F1 score and index of balanced accuracy. Using the resampling algorithm in conditions of the class imbalance, it was possible to increase the reliability of the classification of anomalous bitcoin transactions in comparison with the best known result on the Elliptic dataset.