# Standardize/normalize power law distribution for machine learning

If my data follows a normal distribution I can standardize it for a machine learning algorithm, e.g. logistic regression, by subtracting the mean and dividing the result by the standard deviation.

But what should I do with data that is power law/Pareto distributed, e.g. words in a corpus? How should data in this case be standardized? Or should I prefer simple scale normalizatiom?

-
 Side tip: tree-based ML algorithms does not require transformation at all and most kernel-based ML codes have pretty clever normalization built-in. – mbq♦ Sep 3 '12 at 21:11