Combining Sentiment Analysis with Crypto: Enhancing Machine Learning Models for Price Forecasting with Social Media Insights

For stakeholders and investors, precise price forecasting is becoming more and more important as the market for digital currencies continues to exhibit dynamic and often unpredictable behavior. To improve bit coin price forecasts, this study combines machine learning techniques with data from social media sites. Using the vast and current datasets available through these sites, this research aims to identify patterns, sentiments, and indicators that improve the accuracy of Bitcoin price forecasts. We found that adding the sentiment feature consistently led to superior results when evaluating the models' performance with and without the Twitter sentiment data, with Twitter-RoBERTa-based sentiment generating an average F1 score of 95.3. With an F1 score of 97.3, the most accurate model was an optimized Multi Modal Fusion predictor utilizing sentiment based on Twitter-RoBERTa. By enhancing the precision and resilience of machine learning (ML) methods for anticipating market movements, this work advances the area of financial forecasting and it gives traders, brokers, and investors a useful tool for making well-informed decisions. It accomplishes this by incorporating on-chain data, showcasing the possibilities of sentiment analysis in social media, and practically applying a multimodal fusion model.