On the optimal selection of input data timeframe for deep learning approaches in Bitcoin price forecasting with big data

With the growing technological advancements and inclusion of digital currencies in recent decades, cryptocurrencies have become among the most controversial investment tools. In recent years, Bitcoin, the world’s most popular digital currency, has gained significant trading value and become an attractive option for investing. Since bitcoin has always been in sharp fluctuations, investment in this area involves a high risk for investors. Accordingly, there is an urging need for a tool to reduce the risk of transactions in this market. Nowadays, big data and deep learning networks have entered various fields, including analyzing financial time series and bypassing traditional models with their performance. In this study, an attempt has been made to investigate the efficiency of these networks in the digital currency market using long short-term memory (LSTM) and gated recurrent units (GRUs). The price movement of Bitcoin was predicted in three time intervals: 5, 15, and 30 min. We also compared the efficiency of these two methods and examined the effect of choosing different time frames for the input data of these networks. According to the results, the GRU network outperforms LSTM in most of the studied cases. Also, selecting input data with a smaller time frame can improve network accuracy significantly.

On the optimal selection of input data timeframe for deep learning approaches in Bitcoin price forecasting with big data Keywords:

Deep Learning , Cryptocurrency , Bitcoin , Long Short Term Memory Network , Gated Recurrent Unit Network , Recurrent Network