255
Bitcoin Price Forecasting
The average difference and efficiency of the selected models are computed as fol
lows from Table 15.2 for the year 2017.
The Average Difference for the Year 2017:
For RNN+LSTM Model
: $ 27.881
For RNN+ARIMA Model
: $ 17.122
The Efficiency Produced by the Model for the Year 2017:
For RNN+LSTM Model
: 97.23687%
For RNN+ARIMA Model
: 98.80128%
Overall Improvement
: 1.56448%
The variation in accuracy obtained for both models is depicted as a graph in
Figure 15.3.
From the selected two models, RNN with ARIMA shows better accuracy of
2.078011% compared with RNN with LSTM. The proposed hybrid model RNN
with ARIMA is working as expected and producing outputs with improved effi
ciency of 2.078011% over the previous similarly available model, which is shown in
Figure 15.3. Surprisingly, the combination of stochastic and neural networks works
better in combination and produces better accuracy.
15.6 CONCLUSION
The main objective of this model is to improve the existing method of predicting the
bitcoin price by applying and adding recurrent loops in the activation function inside
the RNN model with ARIMA. The advancements in machine learning techniques
91
92
93
94
95
96
97
98
99
100
101
AUG
30,2016
SEP
15,2016
SEP
30,2016
OCT
15,2016
OCT
30,2016
NOV
15,2016
NOV
30,2016
DEC
15,2016
DEC
30,2016
Bit coin price accuracy in Percentage
Bitcoin price over dates in 2016
Bit-coin price predic on accuracy using RNN+LSTM and RNN+ARIMA
RNN+LSTM
RNN+ARIMA
FIGURE 15.3 Bitcoin price prediction accuracy using RNN+LSTM and RNN+ARIMA.