Modeling and Prediction of Cryptocurrency Prices Using Machine Learning Techniques

Abstract

With the introduction of Bitcoin in the year 2008 as the first practical decentralized cryptocurrency, the interest in cryptocurrencies and their underlying technology, Blockchain, has skyrocketed. Their promise of security, anonymity, and lack of a central controlling authority make them ideal for users who value their privacy. Academic research on machine learning, Blockchain technology, and their intersection have increased significantly in recent years. Specifically, one of the interest areas for researchers is the possibility of predicting the future prices of these cryptocurrencies using supervised machine learning techniques. In this thesis, we investigate their ability to make one day ahead price prediction of several popular cryptocurrencies using five widely used time-series prediction models. These models are designed by optimizing model parameters, such as activation functions, before settling on the final models presented in this thesis. Finally, we report the performance of each time-series prediction model measured by its mean squared error and accuracy in price movement direction prediction.