Hybrid Models for Predicting Cryptocurrency Price Using Financial and Non-Financial Indicators

Cryptocurrency has become very popular and widely used by major businesses as digital currency for online investments and services. However, the price prediction of such digital currencies as Bitcoin and Ethereum is challenging. It involves financial indicators and nonfinancial indicators, such as historical data and social media data, respectively. In this paper, we propose deep learning and hybrid models that effectively incorporate both types of indicators and introduce the optimal algorithms for long-term price prediction of Bitcoin and Ethereum. We conduct extensive experimental evaluations on real data we extracted from financial dataset comprising Yahoo Finance data, and non-financial data consisting of Google Trends data and approximately 30 million related Bitcoin and Ethereum. Our experimental results show that the hybrid models involving LSTM/1D-CNN with ARIMA/ARIMAX outperformed the individual models for the long-term prediction of cryptocurrency prices.