0
$\begingroup$

I am using R software to fit different type of models such as gaussian process model, linear mixed effect model, ANN and etc on many independent variables to estimate my dependent parameter. I have question: is it necessary to normalize my input data such as temperature (use any normality test) before fitting linear mixed effect models or other type of models? In fact, if my input parameters don't be normal, should I use normal function to convert them into the normal shape and then fit the models?? Thanks

$\endgroup$
0

1 Answer 1

0
$\begingroup$

If by "normalization" you mean standardize (i.e. mean 0 with a standard deviation of one, ala scale(x)), then I would say that it isn't necessary, but it is often helpful for at least two reasons.

First, it puts all parameters on the same scale, so that some aren't massive while others are tiny. This is something that can cause trouble with fitting models.

Second, it frequently helps with interpretation of and comparison between each of the estimated parameters.

If you mean something else by normalization (e.g. taking the log of skewed variables), then I'm not completely sure, but I suspect that the utility of doing this will depend on the type of model and the nature of the problem.

$\endgroup$
1
  • $\begingroup$ Thanks, I mean the second. using something like "Log" function to normal skewed input parameters and then fitting the models. $\endgroup$
    – saeed
    Mar 1, 2018 at 20:32

Not the answer you're looking for? Browse other questions tagged or ask your own question.