Predictive Model with Twitter Data for Predicting the Price Volatility of Cryptocurrencies Using Machine Learning Algorithms

Abstract

Cryptocurrency trading is gaining momentum and is one of the fastest growing trading segments in the world. But cryptocurrency prices are highly volatile. This study aims to develop a predictive model to predict the price of cryptocurrency using machine learning algorithms by analyzing Twitter sentiments. This study differs from other studies by providing the tweets as input to the machine learning algorithms and analysing the important words in the tweets that influence the price volatility of cryptocurrencies. Since Bitcoin is the most widely used cryptocurrency in the market, the price of bitcoin is considered for the study. The study makes use of cryptocurrency price and Twitter data from April 4th, 2021 to May 3rd, 2021. The Twitter data is extracted and converted to a document term matrix and is used as predictor variables. Price volatility is the response variable. Three machine learning algorithms, such as support vector machine, decision tree, and random forest, were used for model building. The hyperparameters of the algorithms were tuned to enhance the accuracy. The three models were evaluated based on accuracy, sensitivity, specificity, area under the curve, and the Kolmogorov–Smirnov test. It was found that random forest provided the best accuracy compared to others. Further, the predictor variables were analysed to understand the important words in the tweets that influenced the price volatility. Due to price volatility, there is uncertainty for investors who want to use cryptocurrency as an investment opportunity. This study is a contribution towards using investor sentiment to predict the price of cryptocurrency.