Predicting millionaires from Ethereum transaction histories using node embeddings and artificial neural nets

Abstract

We demonstrate the possibility of predicting the millionaire status of an externally-owned Ethereum account based on information queried out of the Blockchain and features describing its role in the network based on node embedding techniques. These features are fed into artificial neural nets and used for status prediction within a systematic model selection process. We report key methodological insights about training and comparing cross-validated deep and shallow neural networks that yield up to 82 percent testset accuracy in predicting the millionaire status of account users.