Setting neural network parameters

A neural network uses a portion of the historic data to derive patterns that are then used to predict additional data.

1. Open the Model page of the Forecast editor.
2. Click Add.
   The Forecast model selection page of the Add Model wizard opens.
3. Click Neural network and then click Next.
   The Neural network page of the Add Model wizard opens.
4. Specify a value in the Allowable reduction (%) field.
5. Specify a value in the Allowable increase (%) field.
6. Select the frequency at which you expect to see seasonal effects on the data from the Seasonal frequency drop-down list.
7. Specify how many times the neural network runs the data to train itself in the Training iteration drop-down list.
8. Specify the type of neural network algorithm to be used.
   Available algorithms include:

   • Gradient descent - A first-order optimization algorithm.
   • Quick prop - A slight variation of the standard backpropagation of error algorithm.
   • R-prop - A resilient backpropagation of error algorithm.
9. Select the number of hidden neurons from the Hidden neurons drop-down list.
10. Specify the random seed of the neural network in the Random seed field.
11. Specify a value in the Training set (%) field.
12. Select the Show training progress check box to display the training progress of the neural network.
13. Click Finish.
    The Add Model wizard closes and the forecast model is displayed in the table.