2
$\begingroup$

Many of my features are highly skewed as you can see below in the figure. Should I be using such skewed data for modeling purposes? If I cannot, then is there any way to integrate such features in my model? Highly Skewed Feature Since most of my features have a distribution plot like this, can I make a good machine learning model from such data?

$\endgroup$
  • $\begingroup$ How many distinct values? Exact zeros? Can you show a plot of its log (or maybe of $\log(x+1)$? $\endgroup$ – kjetil b halvorsen Jul 24 at 17:30
2
$\begingroup$

Appropriate question.

The added value of preprocessing depends on the type of classifier you will train. If you use nonparametric classifiers like C4.5 (ID3), CART, the multinomial classifier, the webservice insight classifiers, random forests or the like - transformation of your skewed feature values is unnecessary. Their algorithms use histogram-like criteria to choose the optimal classifier parameters.

Classifiers like (deep) neural networks, discriminant analysis, support vector machines, logistic regression - they all use some sort of (local) distance measure. For such models a log-transformation or a power transformation (e.g. $\sqrt{x}$) are highly recommended for your use case.

| cite | improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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