Detecting Malicious Accounts showing Adversarial Behavior in Permissionless Blockchains

Abstract

Different types of malicious activities have been flagged in multiple\npermissionless blockchains such as bitcoin, Ethereum etc. While some malicious\nactivities exploit vulnerabilities in the infrastructure of the blockchain,\nsome target its users through social engineering techniques. To address these\nproblems, we aim at automatically flagging blockchain accounts that originate\nsuch malicious exploitation of accounts of other participants. To that end, we\nidentify a robust supervised machine learning (ML) algorithm that is resistant\nto any bias induced by an over representation of certain malicious activity in\nthe available dataset, as well as is robust against adversarial attacks. We\nfind that most of the malicious activities reported thus far, for example, in\nEthereum blockchain ecosystem, behaves statistically similar. Further, the\npreviously used ML algorithms for identifying malicious accounts show bias\ntowards a particular malicious activity which is over-represented. In the\nsequel, we identify that Neural Networks (NN) holds up the best in the face of\nsuch bias inducing dataset at the same time being robust against certain\nadversarial attacks.\n