A generalized linear model maps a linear transformation of features to some response through monotonic function, does GLM feature selection always go by analyzing the coefs of this linear transformation? Is there any way to go non linear feature selection rather linear feature selection in GLM? with linear feacture selection i've meant to use linear transformation belongs to a GLM model to go feature selection.

For example:

$p(x)=\displaystyle\frac{1}{1+e^{-w.x}}$, where $w=(w_0,w_1...,w_p)$ and $x=(1,x_1,...,x_p)$

in this model linear feature selection can go analyzing the coef's $w_i$ assigned to predictors $(1,x_1,...,x_p)$ and since p(x) is monotonic the $k$ most important variables in $-w.x$, larger absolute weights, are the same as $p(x)$.

With non linear feature selection i mean if there is any feature selection in GLM could go rather use $w.x$ by itself but without explicitly switch $w.x$ to another transformation in model.

  • 1
    $\begingroup$ Please focus your post on a single question. If it's the second, then also tell us what you mean by "non linear feature selection:" your intention is unclear. $\endgroup$ – whuber May 28 at 21:45
  • $\begingroup$ @whuber please look through again sir. $\endgroup$ – Davi Américo May 28 at 22:05
  • $\begingroup$ Here is a paper that actually uses the phrase "non linear feature selection": journals.plos.org/plosone/article?id=10.1371/… $\endgroup$ – kjetil b halvorsen May 29 at 15:40

Your Answer

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

Browse other questions tagged or ask your own question.