Leveraging tabular GANs for malicious address classification in ethereum network

The popularity of ethereum for cryptocurrency transactions attracts malicious actors to engage in illegal activities like phishing, ponzi, and gambling. Previous studies have focused mainly on phishing due to the large number of phishing addresses. However, there is no work done on ponzi or gambling classification due to the limited availability of these addresses, which makes their classification more challenging. In this paper, we propose a machine learning (ML) based method for classifying malicious addresses in ethereum, with a specific focus on phishing, ponzi, and gambling addresses. We use a selective upsampling technique through the tabular generative adversarial network (GAN) to solve limited data problems. We perform not only binary but also multiclass classification on various feature extraction methods, including Trans2Vec and Node2Vec, using Ethereum transactional data. We evaluate our method on $F_1$ score, precision, recall, and accuracy. Our results show that the proposed method is effective in ponzi and gambling detection when compared with the state-of-the-art.