Is it captures the cyclical and trend component in the neural networks models?

In this article, we evaluate the autoregressive neural networks (ARNN) in forecasting time series with trend and seasonal cycle. For the time series analyzed, we found that the combined application of simple and seasonal differentiation factors does not necessarily contribute to better forecasts than if these are applied separately, and that while the preprocessing of data helps to better forecast does not mean that models are not able to capture both the trend and the seasonal cycle present in the untransformed series. In this regard, both the ARNN as the MLP (used as a benchmark) gave better results than the SARIMA process, which in the case of MLP contradicts what is stated in [15], so the assertion that MLPs do not capture these components applies only in special cases. This conclusion can be extended to ARNN model.