Predicting Cryptocurrency Prices Comparing Different Machine Learning Techniques: A Performance Analysis for Pre and Post COVID-19 Pandemic Periods

Abstract

The main purpose of this paper is to compare the forecasting results of time series machine learning models to predict the cryptocurrencies’ future prices for pre-COVID-19, during COVID-19 and post COVID 19 pandemics. Time series data collected from Yahoo Finance was used for the period between January 2017 till February 2024 as test data and the training data to predict 12 months from March 2024 till February 2025. The author undertook three machine learning models: SARIMA, LSTM and FbProphet for the forecasting analysis. LSTM model performs well in predicting the daily price forecasting as compared to SARIMA and Fb prophet models. Bitcoin is predicted to be in the range of $55000 to $65000 by February 2025. Results show a robust trend of volatility during COVID and post COVID periods and pre COVID period was not volatile resulting to no price movements. Based on the forecasting results post-COVID-19 pandemic, the LSTM model outperforms with better predictions than the other models. The findings also revealed that LSTM-RNN model can significantly increase the predictive power in the studies of deep learning models. This paper contributes to the literature on machine learning and forecasting models and the finding provides unique information while modeling the returns. It also insights on what machine learning model is the best to predict movements in the time series data. The results in this paper are expected to enhance our understanding on the role of machine learning models in forecasting future prices of investment instruments in the market, making it valuable for academics, investors, and policymakers alike.