Enhancing Financial Predictions Based on Bitcoin Prices using Big Data and Deep Learning Approach

Abstract

An increasingly popular alternative investing strategy, trading digital money is gaining traction daily. In terms of technological implementation, Bitcoin is among the most prominent digital currencies. Bitcoin is decentralised and doesn't answer to any government, but that hasn't stopped many investors from trading in it and stimulating  the  economy.  The  purpose  of  this  study  is  to  forecast  the  next  day's bitcoin  price  using  five  separate  statistical  and  ML  methods  and  to  evaluate  and contrast them. The ever-changing cryptocurrency industry, however, makes Bitcoin price   prediction   an   increasingly   important   task.   This   study   examines   the effectiveness of several MLP, RNN, ARIMA, and SVM-based models in predicting Bitcoin prices. When applied to the historical price data, an MLP model turned out to be the most efficient, with an R² at 95.9%, while ARIMA was at 90.31%, SVM at 67.3%, and RNN at only 50.25%. The 60-day evaluation proved the proposed MLP model’s accuracy in capturing short-term  price  movements,  thussupporting  the concept  of  a  good  fit.  This  work  is  then  useful  to  establish  sound  guidelines  in including ML and DL strategies in financial prediction, showcasing MLP model to improve decision-making during turbulence. More improvements can include other market factors and better configurations for improved precision and capacity.

Keywords: bitcoin, cryptocurrency, big data, Deep learning, Predictions, IoT, macroeconomics, Multi-Layer Perceptron, Recurrent Neural Network, AutoRegressive Integrated Moving Average, Support Vector Machine