Forecasting bitcoin price movements with machine learning algorithms

Although the term cryptocurrency has entered the financial world in the past decade, it has overcome the trust concerns that existed in the initial period and took its place in the markets as a trading and investment tool. The fact that cryptocurrencies are an internet-based value, has limited supply, ability to transactions 24/7 and not controlled by any authority makes it an attractive investment tool. There are over a thousand known cryptocurrencies today. The first and most widely used of these cryptocurrencies is Bitcoin. Bitcoin, which has been frequently mentioned in the media due to the volatility in its prices in recent days now become more than an investment tool that can be used for daily needs. In this study, by referring to the concepts of Blockchain and Cryptocurrency, will be examined machine learning algorithms used in Bitcoin-specific price movements forecasts, indicators used for forecasting analysis and methods followed in past academic studies and efficiency findings. After the examinations, in order to compare the performance of machine learning methods with big and small data, analyzes have been made on Bitcoin movements for the past 7 years and 1 year. Performance analysis was conducted on the determination of the price direction with 7 different machine learning methods on the datasets that includes technical indicators, daily Tweet numbers and Google Trends data. Finally, the prediction performances of the trained machine learning methods were measured on a new data set. With these performance analyzes, it is aimed to guide the determination of machine learning methods that can be used in predicting the price direction of new cryptocurrencies that have not yet formed long-term data.