A Comparison of Prediction Capabilities of Bayesian Regularization and Levenberg–Marquardt Training Algorithms for Cryptocurrencies

Abstract

Prediction of any currency is performed for identifying and quantifying uncertainties, estimating their impact on results with real-time market value. In this paper, the Bayesian regularization artificial neural network (BRANN) and Levenberg–Marquardt artificial neural network (LMANN) are compared in terms of their prediction abilities. Four cryptocurrencies like Bitcoin, Bitcoin cash, Litecoin, and Ripple price have been taken for comparing their prediction capabilities. The BRANN and LMANN are found suitable for the prediction because both models are used for prediction of time series data and do not depend on any historical features like trends or seasonality. They provide their prediction based on training data. The network outputs are compared in terms of mean percentage error. It is found by experiment that BRANN gives less error than LMANN for large size data. But the performance of both neural networks is less same for small size data.