Automatic time series forecasting using nonlinear autoregressive neural network model with exogenous input

This study aims to determine an automatic forecasting method of univariate time series, using the nonlinear autoregressive neural network model with exogenous input (NARX). In this setting, users only need supply series. Then, algorithm sets up appropriate features, estimate parameters in model, and calculate forecasts, without users’ intervention. The used include preprocessing, tests for trends, application first differences. series were tested seasonality, seasonal differences obtained from a successful analysis. These also linearly scaled [−1, +1]. lags hidden neurons further selected through stepwise optimization algorithms, respectively. 20 NARX models fitted different random starting weights, forecasts combined ensemble operator, order obtain final product. proposed was applied real data, its performance compared several available literature. accuracy measured by mean squared error (MSE) absolute percent (MAPE), results showed that outperformed other models.