Intelligent hybrid model of STS-NARX for prediction of bitcoin price

In this study, hybrid approaches are proposed to deal with the complex features of nonstationary and nonlinear behaviour. The proposed model takes into account some features, such as the seasonal component and the trend as an input data. However, to the best of the authors’ knowledge, statistical patterns, including trend, seasonality and nonlinear behaviour, have rarely been studied simultaneously in bitcoin data. Therefore, a hybrid of structural time series (STS) and nonlinear autoregressive with exogenous input (NARX) is developed to predict bitcoin closing price. The results reveal that the optimal model is the hybrid STS, combining Local Level with Drift Deterministic Seasonal and NARX, featuring two inputs, one output, two delay values, and ten neurons in a hidden layer. This model demonstrates superior performance with the lowest values across all three measurement errors when compared to a single linear STS in bitcoin price prediction.