4

I am trying to determine the significance of my features in my classification using LASSO. However, I could not find any reference or guideline on doing so. I understand that LASSO mainly works for regressions, however, is there any way or any guideline that I could work around with?

If there is no way for it, is there any other similar method for me to determine the significance of my features? E.g. which features affect the classification the most?

Dan
  • 45,079
  • 17
  • 88
  • 157
Sammy Vellu
  • 139
  • 1
  • 5
  • 1
    This is not a programming question, this is a stats questions and belongs on [stats.se]. – Dan Jan 11 '16 at 09:39

1 Answers1

1

You can use the Lasso or elastic net regularization for generalized linear model regression which can be used for classification problems.

[B, FitInfo] = lassoglm(data,group,'binomial','CV',10);
minpts = find(B(:,FitInfo.IndexMinDeviance)); minpts'

Here data is the data matrix with rows as observations and columns as features. group is the labels. minpts will have the list of important features.

Ref : https://in.mathworks.com/help/stats/lassoglm.html

prashanth
  • 4,197
  • 4
  • 25
  • 42