Applies to:
NeuralTools 6.x and Up

 

Question

Is there anything I can do to get a better network so that I’ll have higher confidence in predictions?

Answer

Yes. There are several ways to improve your network’s quality and predictive confidence. Choose the options below that best fit your situation:

Ways to Improve Network Performance

1. Gather more cases
The more unique (non-duplicate) cases you provide, the better NeuralTools can train and generalize a network.

2. Train longer
If you set a fixed training time and NeuralTools did not finish training, increase the allowed training time and retrain the network.

3. Use Best Net Search
If you specified a particular network type, a different type may perform better.
To try this:

• Open the Training dialog

• Go to the Net Configuration tab

• Change Type of Net to Best Net Search

• Retrain the network

This takes longer but often produces better results.

4. Exclude some variables
Removing low-impact or irrelevant variables can speed up training and improve results.
See Eliminate Variables Based on Impact Analysis for guidance.

5. Adjust the training/testing split
If you only have a training dataset and are reserving a percentage of cases for testing, try a different percentage.
The Testing Sensitivity command (Utilities menu) can help you determine the best split.