Empirical Forecasting Analysis of Bitcoin Prices: A Comparison of Machine learning, Deep learning, and Ensemble learning Models

Authors

  • Nrusingha Tripathy Department of Computer Science and Engineering, Siksha ‘O’ Anusandhan Deemed to be University, Bhubaneswar, India https://orcid.org/0000-0002-0272-7479
  • Sarbeswara Hota Department of Computer Application, Siksha ‘O’ Anusandhan Deemed to be University, Bhubaneswar, India
  • Debahuti Mishra Department of Computer Science and Engineering, Siksha ‘O’ Anusandhan Deemed to be University, Bhubaneswar, India
  • Pranati Satapathy Department of IMCA, Utkal University, Bhubaneswar, Odisha, India
  • Subrat Kumar Nayak Department of Computer Science and Engineering, Siksha ‘O’ Anusandhan Deemed to be University, Bhubaneswar, India

DOI:

https://doi.org/10.32985/ijeces.15.1.3

Keywords:

Bitcoin, Cryptocurrency, Arima, LSTM, Prophet, XGBOOST, Prediction

Abstract

Bitcoin has drawn a lot of interest recently as a possible high-earning investment. There are significant financial risks associated with its erratic price volatility. Therefore, investors and decision-makers place great significance on being able to precisely foresee and capture shifting patterns in the Bitcoin market. However, empirical studies on the systems that support Bitcoin trading and forecasting are still in their infancy. The suggested method will predict the prices of all key cryptocurrencies with accuracy. A number of factors are going to be taken into account in order to precisely predict the pricing. By leveraging encryption technology, cryptocurrencies may serve as an online accounting framework and a medium of exchange. The main goal of this work is to predict Bitcoin price. To address the drawbacks of traditional forecasting techniques, we use a variety of machine learning, deep learning, and ensemble learning algorithms. We conduct a performance analysis of Auto-Regressive Integrated Moving Averages (ARIMA), Long-Short-Term Memory (LSTM), FB-Prophet, XGBoost, and a pair of hybrid formulations, LSTM-GRU and LSTM-1D_CNN. Utilizing historical Bitcoin data from 2012 to 2020, we compared the models with their Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE). The hybrid LSTM-GRU model outperforms the rest with a Mean Absolute Error (MAE) of 0.464 and a Root Mean Squared Error (RMSE) of 0.323. The finding has significant ramifications for market analysts and investors in digital currencies.

Downloads

Published

2024-01-08

Issue

Section

Original Scientific Papers