Rug-pull malicious token detection on blockchain using supervised learning with feature engineering

The rapid development of blockchain and cryptocurrency in the past decade has created a huge demand for digital trading platforms. Popular decentralised exchanges (DEXs) such as Uniswap and PancakeSwap were created to address this market gap, facilitating cryptocurrency exchange without intermediaries and hence eliminating security and privacy issues associated with traditional centralised platforms. This, however, due to lack of regulation, results in the emergence of a host of damaging investment fraudulent schemes, including Ponzi, honey pot, pump-and-dump, and rug-pull.In this study, we aim to investigate the problem of detecting rug-pull on Uniswap using supervised learning. We aggregate a list of 23 features and propose the use of a hybrid feature selection technique to find the most relevant features for rug-pull. The classifier, using this refined set of features, outperforms the classifier in the previous studies and achieves an f1-score of 99%, a precision of 97% on non-malicious tokens, and a recall of 99% on malicious tokens. Additionally, we show that the XGBoost classifier, built using these proposed features, can distinguish scam tokens and newly listed tokens, which are often harder to differentiate as they have similar characteristics, and also propose a validation method.