The world of digital money, especially cryptocurrencies are becoming increasingly popular nowadays. These currencies vary from the traditional money or currencies in many ways, such as the various factors responsible for rate trends and changes. One of these factors is the image of the currency in society. Based on this presumption, I explore the possibility to optimize currency rate forecast using data derived from public opinion.
In my thesis I examine the exchange rate forecasting efficiency of the 3 cryptocurrencies with the highest market share. For my researches I will use data derived either from exchange rates and from one of the most popular community pages, Twitter. I use several data mining algorythms to observe exchange rate shifts ot trends. The data derived from Twitter posts and exchange rates will function as parameters for these models. I test the forecasting efficiency of the models by combining several input parameters. In the next step I compare the efficiency rates to select the parameters of the most efficient model. Using this model, I preform a real life analysis to observe gains, advantages, loss and disadvantages generated.
Through the evaluation of the results of my research it had become clear that the K-nearest neighhbour based model is less efficient in rate forecasting than the one based on logistic regression or gradient tree boosting. By analysing the results from using different parameters it is shown that the statistics derived from Twitter data had not improved the forecasting efficiency of the models. However, by analysing emotional effects or using TF-IDF text mining method, the forecasting efficiency can be improved. With results of the real life analysis it can be stated that models using Twitter data are proved to be efficient.