Bitcoin is the most widely known blockchain, a distributed ledger that records an increasing number of transactions based on the bitcoin cryptocurrency. New bitcoins are created at a predictable and decreasing rate, which means that the demand must follow this level of inflation to keep the price stable. Actually, the price is highly volatile, because it is affected by many factors including the supply of bitcoin, its market demand, the cost of the mining process, as well as economic and political world-class news. In this work, we illustrate a novel approach for bitcoin trend prediction, based on the One-Dimensional Convolutional Neural Network (1D CNN). First, we propose a methodology for building useful datasets that take into account social media data, the full blockchain transaction history, and a number of financial indicators. Moreover, we present a cloud-based system characterized by a highly efficient distributed architecture, which allowed us to collect a huge amount of data in order to build thousands of different datasets, using the aforementioned methodology. To the best of our knowledge, this is the first work that uses 1D CNN for bitcoin trend prediction. Remarkably, an efficient and low-cost implementation is feasible due to the simple and compact configuration of 1D CNN models that perform one-dimensional convolutions (i.e., scalar multiplications and additions). We show that the 1D CNN model we implemented, trained, validated and tested using the aforementioned datasets, allow one to predict the bitcoin trend with higher accuracy compared to LSTM models. Last but not least, we introduce and simulate a trading strategy based on the proposed 1D CNN model, which increases the profit when the bitcoin trend is bullish and reduces the loss when the trend is bearish.

CNN-based multivariate data analysis for bitcoin trend prediction / Cavalli, Stefano; Amoretti, Michele. - In: APPLIED SOFT COMPUTING. - ISSN 1568-4946. - 101:(2021). [10.1016/j.asoc.2020.107065]

CNN-based multivariate data analysis for bitcoin trend prediction

Stefano Cavalli;Michele Amoretti
2021-01-01

Abstract

Bitcoin is the most widely known blockchain, a distributed ledger that records an increasing number of transactions based on the bitcoin cryptocurrency. New bitcoins are created at a predictable and decreasing rate, which means that the demand must follow this level of inflation to keep the price stable. Actually, the price is highly volatile, because it is affected by many factors including the supply of bitcoin, its market demand, the cost of the mining process, as well as economic and political world-class news. In this work, we illustrate a novel approach for bitcoin trend prediction, based on the One-Dimensional Convolutional Neural Network (1D CNN). First, we propose a methodology for building useful datasets that take into account social media data, the full blockchain transaction history, and a number of financial indicators. Moreover, we present a cloud-based system characterized by a highly efficient distributed architecture, which allowed us to collect a huge amount of data in order to build thousands of different datasets, using the aforementioned methodology. To the best of our knowledge, this is the first work that uses 1D CNN for bitcoin trend prediction. Remarkably, an efficient and low-cost implementation is feasible due to the simple and compact configuration of 1D CNN models that perform one-dimensional convolutions (i.e., scalar multiplications and additions). We show that the 1D CNN model we implemented, trained, validated and tested using the aforementioned datasets, allow one to predict the bitcoin trend with higher accuracy compared to LSTM models. Last but not least, we introduce and simulate a trading strategy based on the proposed 1D CNN model, which increases the profit when the bitcoin trend is bullish and reduces the loss when the trend is bearish.
2021
CNN-based multivariate data analysis for bitcoin trend prediction / Cavalli, Stefano; Amoretti, Michele. - In: APPLIED SOFT COMPUTING. - ISSN 1568-4946. - 101:(2021). [10.1016/j.asoc.2020.107065]
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11381/2886665
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 66
  • ???jsp.display-item.citation.isi??? 55
social impact