1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

Why statistical model significance test seldom seen in feature selection of Machine learning?

Discussion in 'Education' started by Shaowu, Oct 8, 2018.

  1. Shaowu

    Shaowu Guest


    What I learned from the regression class in statistics told me that in order to remove a feature, or several features, we need to run a statistical significan test by checking the $F$-test related to explained variance by extra parameters and see if they are redundant.

    However, in machine learning community, people tend to use LASSO or other penalty based on $l_1$ regularization.


    I can see LASSO, or optimization-style are more direct in providing the final result, and it finds all at once, of them instead of repeated, multiple-steps, ANOVA-like, statistics-style method.

    But is there any advantage of the later ones?

    Login To add answer/comment

Share This Page