A Comparative Analysis of Ensemble-Based Models for Predicting Cryptocurrency Price Movements

Abstract

This study, "A Comparative Analysis of Ensemble-Based Models for Predicting Cryptocurrency Price Movements," evaluates ensemble machine learning models bagging, boosting, and stacking to improve cryptocurrency price prediction accuracy. Using historical data, models like Random Forest, Gradient Boosting, and Stacking were tested, with Stacking emerging as the top performer (81.80% accuracy, 81.49% F1-score, 88.43% AUC-ROC), outperforming traditional methods like Naive Bayes and Decision Trees. The Boosting Combined model also showed strong results. The research highlights the effectiveness of ensemble techniques in handling cryptocurrency market volatility, offering valuable insights for traders and investors. It underscores the potential of advanced feature engineering and real-time testing to further enhance predictive accuracy, advancing financial decision-making and risk management in the cryptocurrency sector.