Ensemble Machine Learning Techniques for Bitcoin Crypto-Currency Price Forecasting

Abstract

Bitcoin and other Crypto-Currency Price Prediction has been a concern for many financial analytics and business owners. This becomes critically important due to the volatile nature of the Bitcoin. This paper focuses on predicting the next-day Bitcoin price predictions using historical OHLC data from 2019 to 2024 using eleven machine learning algorithms. We have applied fourteen technical features including moving averages, volatility indicators, and lag variables to capture market statistics and price fluctuations. Models that were used in this paper include linear methods (Linear, Ridge, and Lasso regression), ensemble techniques (Random Forest, XGBoost, Gradient Boosting, AdaBoost), instance-based learning (KNN), support vector machines (SVR), decision trees, and deep learning (LSTM networks). Robust performance is ensured by the Five-fold cross-validation. The results clearly show that Lasso regression outperforms other algorithms with a RMSE of $727.33 and R2of 0.971, achieving superior performance in comparison to complex ensemble methods.