AN APPLIED STUDY OF RNN MODELS FOR PREDICTING CRYPTOCURRENCY PRICES

Abstract

In recent years, cryptocurrency, and its foundational technology, blockchain, have become increasingly popular.As the value of some cryptocurrencies has cataclysmically risen and fallen, many people are curious about what the future of cryptocurrency prices look like, and what means are available to predict that future.We conducted experiments on three RNN models to determine whether one performed better than the others when predicting the price of Bitcoin, Ripple, and Litecoin.We also collected Google trends data for these three cryptocurrencies and ran additional experiments to explore whether this additional data significantly boosted the prediction accuracy of these models.After performing all our experiments, we found that among the three RNN models we tested, none performed significantly better than the others.Additionally, we concluded that supplementing the three models with Google trends data did not significantly boost the prediction accuracy of the models.We discuss implications of this research in the paper.