A new method of ensemble learning: case of cryptocurrency price prediction

This work proposes a novel method of ensemble learning for time series prediction. Different machine learning-based models have been integrated, and a combined prediction model has been created. The objective of the ensemble model is that it must outperform all other individual models that are used to construct the ensemble model in terms of producing excellent predictions. The field of cryptocurrencies has been selected as the domain of this work where the focus is to predict the cryptocurrency prices using the proposed model. A new regression model is proposed and implemented in this work. Different machine learning techniques have been adopted and integrated to form a combined prediction model. The machine learning models include deep neural networks, support vector regression, and decision trees. The regression scheme has to be implemented on each machine learning model separately as well as their performance is also to be improved. The combined prediction model requires optimal weights generation for integration, and therefore, time complexity is a concern. A large set of experiments have been carried out on various cryptocurrencies and the results are displayed. Real-world data has been used here and a comparison is also performed. It is observed that the combined prediction model outperforms other models resulting in excellent predictions capturing most of the nonstationary movements in the data.