A Comparative Study of Time Series Models for Bitcoin Price Prediction

Abstract

Bitcoin is the very first digital or cryptocurrency based on Block chain concept. The objective of this paper is to forecast the price of Bitcoin using various Machine Learning Time Series models like: Moving Averages (MA), Autoregressive Integrated Moving Average (ARIMA), Extreme Gradient Boosting (XGBoost) and Long Short-Term Memory (LSTM). As we know that the price of Bitcoin is very volatile in nature so producing appropriate predictions is difficult. Also we know that Bitcoin Price nature is not stationary, so we have converted our non-stationary data to stationary for models like ARIMA which works properly on stationary data only. At last, we have compared the results of MA, ARIMA, XGBoost and LSTM for Bitcoin prediction based on RMSE and we found that first three models have given somewhere similar results whereas LSTM has given different.