Forecasting Bitcoin Price Trends: Integrating Sentiment Analysis and Machine Learning

Abstract

With the increasing scrutiny of Bitcoin and other cryptocurrencies in recent years, there is a growing significance in studying methods to predict their values, including sentiment analysis on Twitter-collected data. The primary objective is to perform sentiment analysis on Bitcoin prices and examine the predictive accuracy of various machine learning algorithms. The approach cleans up Twitter data that is relevant to Bitcoin. Followed by quantifying people’s opinions on Bitcoin using sentiment analysis. With this data and Bitcoin’s price history, a binary target variable for price movement is built to specify the price variations. In order to predict the changes in Bitcoin price, several machine learning models are developed and evaluated. Among these models are Long Short-Term Memory networks, Decision Trees, Random Forests, and Linear Discriminant Analysis are mostly applied. The proposed approach is the enhanced Random Forest base classifier to improve the prediction accuracy. With an F1-score of 64% and 70% accuracy, the Enhanced Random Forest model performs better than the existing models. This model illustrates how sentiment analysis and machine learning may be combined to forecast cryptocurrency prices.