Hi Ikenna,

You are able to use a variety of normalization forms in your research. I think the common normalization is zero-score, that is (x-mean)/variance. In addition, you can also make use of the form (X-Xmin)/(Xmax-Xmin). Maybe you do not need normalization method in your problem. So you need to try which one is the best for your issue.

I don't know your issue clearly since I am not familiar with Prediction of carbonation depth.

Dear Ikenna Uwanuakwa,

In my opinion, if you have different features when they have different ranges it's better to normalize columns independently.

Best Regards...

Hi Ikenna,

Please check the following links:

http://stats.stackexchange.com/questions/70801/how-to-normalize-data-to-0-1-range

http://www.benetzkorn.com/2011/11/data-normalization-and-standardization/

keep in your mind that you have to normalize features with different ranges independently as mentioned previously.

thank you all

Try with Z-score or Robust Z-score normalization.

Still simpler method is divide the values of attribute or feature with its own maximum value., i.e Xi (normalized) = xi / Max{x}

NORMALIZATION OF DATA IS INCORPORATED TO REMOVE BIASNESS. DATA CAN BE NORMALIZED IN THE RANGE -1 TO +1 OR 0 TO +1. IF TANGENT HYPERBOLIC ACTIVATION FUNCTION IS USED IN HIDDEN LAYERS, THE NORMALIZATION OF DATA IN THE RANGE -1 TO + 1 WOULD BE BENEFICIAL AND PROVIDES THE DESIRED SENSITIVITY TO THE NEURAL NETWORK WHEN SUBJECTED TO SMALL CHANGES IN THE INPUT VALUES. DATA ENTERED IN COLUMNS RELATED TO DIFFERENT ENTITIES SHOULD BE NORMALIZED INDEPENDENTLY. 

Without normalizing inputs to the ANN the following codes works well for any type of problem with single output feature.

https://www.researchgate.net/publication/305325563_Codes_in_MATLAB_for_Training_Artificial_Neural_Network_using_Particle_Swarm_Optimization

Code Codes in MATLAB for Training Artificial Neural Network using...