CryptoCurrency Time Series analysis : Comparative analysis between LSTM and BART Algorithm

Background: Cryptocurrency is an innovative digital or virtual form of money thatuses cryptographic techniques for secured financial transactions within a decentralized structure. Due to its high volatility and susceptibility to external factors, itis difficult to understand its behavior which makes accurate predictions challengingfor the investors who are trying to forecast price changes and make profitable investments. Time series analysis is a crucial technique that analyzes historical pricedata, trading volumes, etc., and anticipates future price trends behavior. which isone of the most precise and effective techniques for the prediction of prices and alsofor predicting trading pairs. So in this study, we compared machine learning anddeep learning algorithms with time series analysis to find out which algorithm givesaccurate predictions for the trading pairs. Objectives: The objective of this thesis is to identify the efficient algorithm thatproduces accurate predictions of trading pairs using time series analysis using different evaluation metrics like RMSE, MSE, MAE, and R-squared. Methods: SLR has been conducted to choose the algorithms based on the literature survey. The algorithms that we choose here are BART and LSTM to performa time series analysis for forecasting the cryptocurrency trading pair values. The selected data is used for training, validating, and, testing over the selected algorithmsfor the analysis of forecasting future price values. Firstly, the models were importedusing their respective built-in functions and the training data was used for trainingthe created model and the testing data was used for the testing of the model. Next,the time series analysis is implemented using the ARIMA model, the actual valuesand the predicted values over the implemented analysis were examined through thegraphical analysis. After this, the evaluation metrics are calculated on each of theimplemented algorithms to find out which has shown better results in predicting future values. Results: In this section, after the implementation of the method, it is found thatthe BART had shown higher performance compared to LSTM. We can observethat the RMSE of the LSTM algorithm is (0.012429) and BART the algorithm is(0.00372), MSE for LSTM is (0.000154) and for BART is (0.0000138) MAE forBART is (0.00023) and for LSTM is (0.0034) whereas R-Squared values for LSTMis (0.2925) and for BART is (0.9364). Conclusions: This study uses the ARIMA model to forecast cryptocurrency tradingpairs and compares its effectiveness to that of the BART and LSTM algorithms using evaluation metrics like RMSE, MSE, MAE, and R-Squared in particular, BARToutperforms the LSTM and exhibits better values for all the four evaluation metrics