Abstract

Long-term research has been conducted into the issue of anomaly detection in Bitcoin transactions, and a lot of different approaches based on network analysis have been put up as potential solutions.

Although there are several outcomes that look to have quite a bit of potential, none of the methods appears to be definitively better than the others. The Elliptic dataset was used in this study and various machine learning algorithms were tested and graph analyses on Bitcoin transactions in order to classify Bitcoin transactions as either licit or illicit. To deal with the evident class imbalance in the illicit class on the dataset. the model is trained using Autoencoder. After that, several machine learning classifiers were put through their paces, and four of them emerged victorious. Random Forest, Logistic Regression, XGBoost, and Multi Layered Perception were shown to have the greatest overall performance when compared to the other classifiers. These are the top four classifiers.

Anomaly Detection in Bitcoin Network Using Classification Models