Forecasting cryptocurrency prices time series using machine learning

Abstract

This paper describes the construction of the short-term forecastingmodel of cryptocurrencies prices using machine learning approach. Themodified model of Binary Auto Regressive Tree (BART) is adapted from thestandard models of regression trees and the data of the time series. BARTcombines the classic algorithm classification and regression trees (C&RT) andautoregressive models ARIMA. Using the BART model, we made a short-termforecast (from 5 to 30 days) for the 3 most capitalized cryptocurrencies: Bitcoin,Ethereum and Ripple. We found that the proposed approach was more accuratethan the ARIMA-ARFIMA models in forecasting cryptocurrencies time seriesboth in the periods of slow rising (falling) and in the periods of transitiondynamics (change of trend).