Cryptocurrency Price Prediction

Abstract

Cryptocurrencies are a digital currency which is safeguarded by cryptography. With emerging popularity and trust towards this technology, cryptocurrencies get high priced and the public starts to invest in them. In order to generate a significant return on investment, this project aims to create a web application that helps users to predict cryptocurrency prices. Machine Learning methods like Long Short Term Memory(LSTM) Network and Statistical Methods like Exponential Smoothing, Vector Autoregression are used to predict cryptocurrency prices. However, these methods did not take social media perception as a consideration, we will implement sentiment analysis and feed the output as an input of the models and investigate any improvement in the accuracy.

Models are all composed and investigated through experiment, with 14 days investigation, the models are not sensitive to a huge change. For the new approach, which is the LSTM Network with sentiment analysis, the performance was not favorable but some intuitions were introduced. The sentiment analysis scores may affect the final results by: i) Unrelated posts, 2) Subjectivity of a post, 3) Grouping sentiment values. Suggestions which may improve prediction results are introduced for future work.