Bitcoin estimation and prediction through neural network and machine learning

Abstract

Bitcoin is the most popular purely digital cryptocurrency nowadays. It started in 2009 as a peer to peer payment system, decentralized, pseudo-anonymous and secure system of money. The main objective of this study is to estimate and predict the weekly close bitcoin price by including other variables like commodities, indexes and demand / supply variables through different kind of machine learning and deep learning such as multiple regression, time series ARIMA model, artificial neural network, combination of multiple regression and ARIMA model, and finally combination of multiple regression, ARIMA model and artificial neural network. The results show that the combination of multiple regression, Time Series ARIMA model, and artificial neural network is the most accurate model with an MAPE = 0.85% for the short term prediction of the close price of Bitcoin.