Machine Learning Based Framework for Cryptocurrency Price Prediction

Abstract

Cryptocurrency, a digital currency, acts as a medium of exchange through the Internet. The main agenda behind cryptocurrency being so popular these days is the desire for reliable, long-term value without the involvement of any central authority like banks. The power lies in the hands of the currency holders which resolve the problems of the traditional currencies by adopting a decentralized system. Predicting the future price of different cryptocurrencies is a prominent area of interest for individuals or investors. In this work, we use a dataset collected from the coinmarketcap website for the duration of September 2014 to March 2022. The outcome of this work is compared to the existing algorithms for time series data analysis namely the Auto Regressive Moving Average Model (ARIMA), FbProphet, and several ensemble models on the basis of their accuracy in predicting the future price. We also create different ensemble frameworks for the prediction of the cryptocurrency price. To form the ensemble models, we initially select the three best-performing regression models on the dataset, namely Extra Trees, Random Forest, and Decision Trees Regressors. Our findings indicate that the ARIMA model performs better than the ensemble model with the lowest RMSE MAE and MSE.