GAN-based Abnormal Transaction Detection in Bitcoin

Abstract

Since its inception, blockchain technology attracts great attention from the industry and academia. With its development, cryptocurrencies such as bitcoin based on blockchain technology gradually emerge and enter the financial field. Meanwhile, malicious behaviors aimed at bitcoin become more and more common and cause huge damage to cryptocurrency users and the evolution of blockchain technology, which prompt researchers to establish various models to deal with this problem. In this paper, we collected the historical bitcoin transaction dataset and extracted features from it. After standardizing features, we used an unsupervised learning model based on Generative Adversarial Networks (GAN) to detect dataset containing more than 30 million normal and 108 malicious samples and reached a precision of 23% and recall value close to 100%.