The normalization is not always advisable. because it hide/reduces some of the data variation. Bt..

If it gives good results even in visualization can not ignor this. You can take and use these normalized values. Normalization is the technique by which You can describe your results, when it is adding positive results in your research.

Roger that Santosh. I'll go that way, through normalization. Thanks for the feedback!

Mr. Moreno,

You should definitely normalize all your datasets. That, especially for your case, in which variables have different natures (units).

Secondly, if you want to best scrutinize and discriminate relationships among variables regardless of their inherent variability, you should also whiten their datasets. That is, treat them so they have unit variance. In fact, this way your PCA plot should better separate data points, and your scores provide you with a more in depth insight as to what you are actually working with as opposed to when you use raw data.

I hope I could help.

Best regards,

Gabriel

That helped indeed Gabriel. 

Dear Bernabé,

I agree with Gabriel and should add that, in most classification and pattern recognition problems, normalization to zero mean and unit variance is used to make features from different nature and different measurement units comparable with one another.

I strongly recommend you to do the feature wised normalization to zero mean and unit variance before you consider any of them against the other one.

All the best

Peyman

You should normalize your data to get better picture of your data and do a better statistical treatment becoz many statistical tests need data to be normalized

Thank you Peyman & Ummer for your answers.

I have proceeded with the normalization of the data-set. The output graphics turn out to be more explicative. Thanks again.

Cheers!

can I ask how did you normalize the data?

Greetings Valentina,

I normalized the data through this formula: (Xn = X – X/SD)

being X: the raw data value; SD: standard deviation of the data group; and Xn: the normalized output of this operation.

Regards

hello!

I think the second X should be the mean if I am right.

thanks for your answer. Would you think that Log transformation will work as well for normalization of the data?

if I am reading correctly the book (Quinn, keough about experimental design), they reccomend to center the data so actually Xn= X-Xmean before the multivariate statistics.

with my data set although I have displayed better the samples and variables when I log10 normalize them.

does it make sense? is there any opinion about this option?

thanks a lot for sharing conversation!

best

You're right about the 2nd X (Xmean) 

Normalization is used to treat similarly variables with different scales and units. 

Log transformation is considered a very strong transformation, sometimes leading to a reduced data variation which can hide some patterns. But, if you've better displayed your samples when log-transformation then you should probably keep on doing it.

Very useful book by the way. I've just downloaded it.

Best