Are there features that the input and target values should follow before training in a neural network?

Even with normalization or mapping in a range [-1,1] before training, I have seen that a good scaling of the data may improve the performance of the neural network. How to define a good scaling for both the input and output data when the range is very large?

Answer 1

[Greg Heath]

You are creating a problem that doesn't exist.

Pretraining standardization can be used to eliminate or modify outliers. Then you can use either or both standardization and/or normalization.

Hope this helps.

Thank you for formally accepting my answer

Greg