Transient Empirical Machine Learning Models based on Bit Coin Price Prediction using High and Low Values

Introduction: Cryptocurrency is the new mode of transaction that is slowly replacing the conventional assets of transactions. With the rise and advancement in the technology called blockchain and the volatile nature of the market, cryptocurrencies are showing possibilities of making huge profits. This new trend is slowly becoming accessible to all classes of society in every part of the world while creating a great opportunity for scholars and analysts to conduct new research. However, unlike the stock market, the prices of cryptocurrencies show highly volatile and dynamic behaviour. Methods: Several research in the past have been performed to predict the nature and other aspects of cryptocurrencies with the aid of machine learning to forecast the future market based on past data. In the proposed work, 6 different machine-learning models have been used for forecasting the prices of cryptocurrencies. The algorithms used are Autoregressive Integrated Moving Average (ARIMA), Gated Regression Unit (GRU), Long Short-Term Memory (LSTM), Bi-LSTM and Vector Autoregression (VAR). Results: Further, the accuracies of different models are compared and analysed. Furthermore, these models are used to predict the first-order difference between 24 hours of all-time high and low values. This evaluation gives us an idea of where the price band of the currency and where is it most likely to be present during any 24-hour timespan. Conclusion: This can effectively lead to better decision making in choosing crypto-currency. The GRU model achieved the best performance in cryptocurrency prediction with an RMSE of 0.2242 and MAE of 0.1598 for BTC prices, while the LSTM followed closely with an RMSE of 0.3473 and MAE of 0.2557. Both models outperformed ARIMA, Bi-LSTM, VAR, and FB Prophet, which showed significantly higher error rates.