Detection Of Illicit Transactions In Bitcoin Based On Illegality Quotient

Detection of illicit transactions refers to discovering unknown illicit transactions,and plays a critical role in anti-financial crimes.Here,illicit transactions refer to transactions involving illicit activities such as financial fraud,ransomware,terrorist organizations.Existing methods detect illicit transactions by extracting features of transactions and building a classification model.The features include number of inputs/outputs,transaction fee and average BTC received/spent by the inputs/outputs,etc.The major problem of these methods is that they try to detect illicit transactions primarily by features.The essence of the problem is that existing detection methods do not extract features which can reflect the illegality of the transaction directly,just use other features to reflect illegality indirectly,leading to limited detection precision and low recall.The paper argues that illegality of a transaction rooted in the illicit activities associated with it.Suppose that there exist two transactions with similar features,the one that involves in an illicit activity is illicit,while the other that does not is licit.Illicit transactions usually associate each other.Based on the above observations,this paper comes up with a notion called Illegality Quotient to measure the degree of illegality of a transaction in a direct way,propose an algorithm to quantify it,and integrate it into the existing models to improve their performance.Another problem for predicting illicit transactions is insufficient illicit transactions serving as samples for building models.Illegality Quotient allows us to choose the transactions with high Illegality Quotient as samples to expand illicit samples.The paper evaluated the proposed approach on models including Logistic Regression(LR),Random Forest(RF),and Graph Convolutional Network(GCN).The experimental results indicate that the F1 score increases by 12.86%on average by integrating Illegality Quotient.The notion called Illegality Quotient and the method of expanding illicit samples can assist to detect the use of Bitcoin for criminal or fraudulent purposes,and improve the detection precision and recall of existing models.