No, they are not equivalent. And you may not need to normalize the sample spaces in matlab nntool.

Dear Gaurav Kumar,

Am not clear with your question. Do you mean to say the prediction error (%) obtained when you normalized the predictors (input variables or data) increased significantly in comparison with the prediction error (%) obtained using the original data (input variables)?

Liming Yuan

What if the sample space is too large and it contains data with big variation between MIN and MAX? Normalizing helps to bring all the values between 0 and 1.

Oluwarotimi Williams Samuel

I meant to ask that when I am normalizing all the input and output data before entering in MATLAB then I am getting good results. But when I am de-normalizing the predicted output and comparing it with original output then the error% is increasing. Is it okay to keep it in normalized form and compare?

Gaurav Kumar

Normalization helps in your case. Try changing the training function and number of neurons to reduce the error. Also, the sample spaces may contain a lot of noise data. you can process the data at the beginning to eliminate the outliers. Or adopt principal component analysis to extract the principal components and use them as the sample spaces.

Dear Gaurav Kumar,

From your explanations, I could deduce that the normalized dataset (Input and output) yielded better results in comparison to the un-normalized output (which I would refer to as the original output). The thing here is that, you may need to compare both results with your expected targets to know exactly the one that has the closest match with your targets. This will help you decide which of the above approaches to adopt in your research provided that the results can be reproduced via the method that gives the best performance.

I hope you would find this tips helpful in your research.

Good luck.

Oluwarotimi Williams Samuel

So is it equally acceptable if I keep the data in normalized form and measure the error?

Because I am getting better results with this method. I can mention in the paper that due to large variations, normalization was done and normalized results were only compared.

Hi Gaurav Kumar,

I think it is okay if you clearly state in your paper that you applied normalization technique to your data in order to obtain more accurate results, If possible, you can also describe the normalization process in detail to give readers proper understanding. Meanwhile, you also need to be sure that your method can be reproduce especially under the same experimental conditions.

Hope you would find this tips helpful.

Good luck