A Time-Series Cryptocurrency Price Prediction Using an Ensemble Learning Model

Abstract

Due to the high volatility in the cryptocurrency market, it is quite challenging to predict the price accurately; therefore, there is a great need for strong prediction models. In this paper, we propose a time-series cryptocurrency trend prediction framework based on a machine learning ensemble learning approach, which combines several machine learning models to achieve higher accuracy and generalisation. Historical prices (including the open, high, low, close, and trading volume) were preprocessed and input into a hybrid LSTM-GBM-RFs ensemble model. The ensemble model combines the merits of individual learners while mitigating their weaknesses through weighted averaging. Through experimental results on Bitcoin and Ethereum datasets, we demonstrate that the ensemble of models outperforms the individual models in terms of MAE and RMSE. This study demonstrates the potential of data fusion for modelling the temporal properties of cryptocurrency time series, paving the way for the further development of real-time decision-making recommendation systems.