Hi Dwaipayan,
In MATLAB's NARNET (Nonlinear Autoregressive Network), the "feedbackDelays" parameter specifies the time steps that the network uses to feed back its own outputs as inputs. It essentially defines the history length that the network considers when predicting future values. By setting "feedbackDelays", you control how many past outputs the network looks at to make its predictions. This is crucial for capturing temporal patterns in time-series data.
For example, if you set "feedbackDelays" to [1:2], the network will use its outputs from the last two time steps as inputs for the current prediction. This helps the network capture temporal dependencies and patterns in time-series data, making it more effective at modeling sequences where past values influence future ones. Adjusting "feedbackDelays" allows you to tailor the network to the specific dynamics of your data.
For more information on NARNET, you can refer to the link mentioned below:
