Decoding Market Emotion from Blockchain Activity: A Data-Driven Sentiment Classifier

TL;DR - A study that classifies Bitcoin market sentiment by fusing on-chain blockchain data, historical price data, and daily Twitter sentiment, showing that combining these signals yields meaningful, interpretable market insights rather than price predictions.

• Merges on-chain transaction metrics, Bitcoin price history, and daily Twitter sentiment classifications into a normalized dataset for sentiment analysis.
• Tested multiple ML models with cross-validation; Gradient Boosting (XGBoost) performed best, achieving an average F1-score of ~0.84.
• Used SHAP (game-theory-based interpretability) to quantify how on-chain features contribute to predictions, improving transparency.
• Framed as explanatory (understanding sentiment) rather than predictive (forecasting prices), with deep learning flagged as future work.