Machine Learning and the Network Analysis of Ethereum Trading Data

Since their conception, cryptocurrencies have captured the public interest, motivating a growing body of research aimed at exploring blockchain-based transactions. This said, little work has been done to draw conclusions from transaction patterns, particularly in the realm of predicting cryptocurrency price movements. Moreover, research in the cryptocurrency sphere largely focuses on Bitcoin, paying little attention to Ethereum, Bitcoin\u27s second-in-line with respect to market capitalization. In this paper, we construct hourly networks for a year of Ethereum transactions, using computed graph metrics as features in a series of machine learning models. We find that regression-based approaches to predicting Ether prices/price deltas primarily and almost exclusively rely on using current prices, motivating the need for classification models to predict price/up down movements rather than raw prices. Across a handful of such classification models, using hourly network metrics as input features, we are able to outperform baseline up/down prediction F-1 scores by up to 14%, accuracy by up to 5%, precision by up to 50%, and recall by up to 7%. These findings have implications for the future of cryptocurrency price prediction and trading activity, and suggest further research