252

Blockchain Technology

Similarly, an experiment was conducted to predict the bitcoin price on different

dates: 11 October 2017, 13 July 2017 and 24 July 2017. The different graphs obtained

using RNN with ARIMA and RNN with LSTM indicate that the variation in the bit­

coin price is not uniform The experiment results also shows that with an unexpected

deep and steep fall in the prices, almost all the time RNN and ARIMA stayed close

to the real value, giving a better prediction, which is more reliable than RNN and

LSTM. Though the RNN and LSTM performs well, it does not match the actual

price of the bitcoin. The improvements of accuracy and efficiency over actual and

other selected models are described in Table 15.1 for the year 2016.

The average difference and efficiency of the selected models are computed as

follows from Table 15.1.

The Average Difference for the Year 2016:

For RNN+LSTM Model

: $ 29.52634

For RNN+ARIMA Model

: $ 4.84223

The Efficiency Produced by the Model for the Year 2016:

For RNN+LSTM Model

: 95.73135%

For RNN+ARIMA Model

: 98.32289%

Overall Improvement

: 2.591542%

Similar calculations were also carried out for the year 2017, which are described in

Table 15.2.

FIGURE 15.2  Bitcoin price prediction on 13 September 2017 – actual price, RNN with

ARIMA and RNN with LSTM.