Exploring the Use of Advanced Machine Learning Techniques to Detect Fraudulent Cryptocurrency Exchanges

Abstract

The decentralized nature, self-regulation, and rapid transaction speeds of the cryptocurrency landscape foster both innovation and fraud. As crypto gains attention, fraudulent exchanges posing as legitimate platforms present significant financial risks. Traditional reporting methods, like those of CryptoScamDB, are becoming less effective. This study addresses the need for automated techniques to identify and classify fraudulent exchanges. We found that features related to domain information, website security, and user reviews are crucial for detection. A comparative experiment using machine learning models—random forest, support vector machine, k-nearest neighbor, light gradient boosting machine, and multilayer perceptron—showed that the light gradient boosting machine (95.9%) and random forest (95.5%) had the highest accuracy. An ensemble model using soft voting achieved a higher accuracy of 96.2%, outperforming its base models. These results demonstrate that machine learning is a viable approach for automating the classification of fraudulent exchanges.