Comparison of Deep Learning Algorithms for “Bitcoin Cash” Price Prediction

Abstract

Price prediction of cryptocurrencies is of great importance for investors and researchers. However, due to the nonlinear nature of the cryptocurrencies, evaluating distinct features of time series data for accurate price prediction is challenging. There are numerous studies on predicting cryptocurrency prices using deep learning algorithms. This article compares some popular deep learning algorithms, including Multilayer Perceptron (MLP), Unidirectional and Bidirectional Recurrent Neural Networks (RNN), Unidirectional and Bidirectional Long Short-Term Memory (LSTM), Unidirectional and Bidirectional Gated Recurrent Unit (GRU), and Convolutional Neural Networks (CNN) for Bitcoin Cash prediction. The results obtained using Mean Absolute Error (MAE) and Mean Squared Error (MSE) indicate that Unidirectional LSTM and Bidirectional LSTM outperform other algorithms in terms of prediction. Therefore, they can be considered as the best algorithms. In addition, according to the comparison made with other papers, the importance of data and hyperparameters is shown in terms of how much they affect the results.