Detecting Anomalous Transaction Patterns on Cryptocurrency Exchanges using On-Chain Data

Exchanges for cryptocurrencies are essential to the cryptocurrency market. However, while offering user services, certain exchanges are suspected of engaging in various strange or malevolent activities, including money laundering and wash trading. Furthermore, it has been noted that these actions frequently result in an unusual rise in the transaction amount. Determining whether and when the anomalous transaction amount happens in the exchange is therefore a topic worthy of investigation. This study gathers a comprehensive dataset of exchanges, and then it does a correlation analysis to identify the key variables influencing the transaction volume of various exchanges. Next, using deep learning, a prediction model of how different parameters affect the transaction amount is obtained. To establish a foundation for the identification of anomalous transaction amounts, the difference between the anticipated and actual transaction amounts is computed. Ultimately, a case study on the detection results reveals that some anomalous transaction amounts are connected to industry events and policy changes, while the remaining quantities are thought to be connected to illicit activity. The root mean square error ranges from 0.07 to 0.7, the mean absolute error ranges from 0.02 to 0.09 and the mean absolute percentage error ranges from 0.3 to 0.5 for all the five crypto exchanges for our prediction if anomalous transaction by deep learning models.