Cryptocurrency Price Prediction: A Comprehensive Analysis of High-dimensional Features with Deep Learning Techniques

Abstract

The most well-known encrypted money, Bitcoin, has a lot of promise in the future. Investors and traders always try to find a technique to forecast cryptocurrency prices to lower their risks and boost profits. However, predicting the price of cryptocurrencies is a difficult undertaking because of their unpredictability, volatility, and mobility. Different prediction architectures have been developed by researchers using machine learning (ML), deep learning (DL), and statistical methods. In this work, predictions are made utilizing the AutoRegressive Integrated Moving Average (ARIMA), Extreme Gradient Boosting (XGBOOST), Gated Recurrent Unit (GRU), and Long Short-Term Memory (LSTM) models. The historical bitcoin market data is chosen, it spans the months of January 2012 through September 2020. The LSTM model performs well when compared to other models, producing a minimal Mean Absolute Error (MAE) of 5.836 and a minimal Root Mean Squared Error (RMSE) of 7.472. Increased return on investment can be achieved by investors by making well-informed decisions on what to buy, hold, or sell. That’s particularly the case with predictions about the price of Bitcoin.