249

Bitcoin Price Forecasting

(Awoke et al., 2021). Sequence learning algorithms are widely applied for time series

forecasting, and LSTM is one of the sequence algorithms. It has not been applied

specifically to cryptocurrencies like Bitcoin, Ethereum, Binance Coin, Dogecoin,

XRP, Tether, Cardano and so on. Hence, a research model has been proposed to pre­

dict the bitcoin price with LSTM and with a variation in LSTM called AR (2). This

model performed well compared with other models in predicting bitcoin prices very

accurately (Livieris et al., 2021).

A survey has been conducted to analyse the impact of different bitcoin price pre­

diction techniques such as statistical techniques, machine learning techniques and

deep learning techniques. The research results indicate that compared with statisti­

cal techniques (which require more assumptions), machine learning techniques per­

form well in predicting bitcoin price (Khedr et al., 2021).

Bitcoin price is also influenced by many factors such as demand raised for bitcoin,

the amount of bitcoin in circulation, and the exchange value of bitcoin. These param­

eters are applied to predict the price of bitcoin using time series analysis methods

like GRU, ARIMA, and LSTM (Gupta & Nain, 2021). The volatile nature of bitcoin

takes time series analysis as a very important model for price prediction. Hence, an

ensemble of time series models combined with a neural network has been developed

for bitcoin price prediction. Two kinds of shifts have been found in bitcoin prices:

Deterministic and moderate. According to a time interval (for example day, hour,

minute, etc.,), the price change is decided and the shift is stated as a moderate or

inevitable (Shin et al., 2021).

From the literature, it has been found that bitcoin is very volatile, and its price

changes very quickly. Moreover, it is one of the decentralized currencies. Hence,

predominant models are required to predict the price of bitcoin very quickly and

more precisely. An alert system should be developed with machine learning when

the threshold level of bitcoin price is reached (Shankhdhar et al., 2021).

A special kind of research has been conducted to investigate the different pre­

diction techniques available to predict the bitcoin price. This research has empha­

sized that predicting bitcoin price with higher accuracy is very important, as it is

essential for investors for investment and good profit-making. Also, it is essential for

policymakers to make policy decisions and for researchers to better understand the

financial market situation and investments. This research has identified that many

statistical models, time series analyses, machine learning techniques and deep learn­

ing techniques are applied to predict the price of bitcoin. When these results are

compared with real bitcoin price analytics, a huge difference is identified. Hence,

the research suggests that more sophisticated and specialized models and algorithms

need to be developed to conduct bitcoin price analytics (Pintelas et al., 2020).

15.3  ARIMA WITH RECURRENT NEURAL

NETWORK FOR FORECASTING

Time series analysis is one of the popular methodologies applied for the predic­

tion of future performance from past performance. From the literature, it is evident

that even with many stochastic models in this game for bitcoin price prediction, it