Use of normalization in a neural net classification.

shadow15jose

Hello, I'm trying to make a classificacion using a data set and a neural network, so I need to normalize data. I'm using a split data to create two sets: one for training and the other one for testing. My questions is: Do I need to normalize data before I split it, or it's better to do after split just for the training set?

lionelderkrikor

Hi @shadow15jose,

 

1.There are operators which are dedicated to the validation in RapidMiner (you don't need to split your dataset) :  

 - Split Validation operator

 - Cross Validation operator (I advise you this operator)

 

2. The normalization is performed by default by the Neural Networks model itself. 

 

 

Regards, 

 

Lionel

Telcontar120

Agreed with @lionelderkrikor that you should be using Cross Validation in almost all circumstances.

As far as normalization goes, if you are only using the base NN operator, as indicated it can perform that automatically.  However if you wish to consider alternative algorithms, you might still want to normalize your data separately, since many other learning operators do not include that option.  So I would still recommend doing your own normalization.  You may also want that later if you build a model and want to apply the same normalization scheme to future records to be scored.