Let's say I have a dataset with 30 variables and 15 target variables , I want to make several models to predict the 15 variable , so ive thought of grouping them and predicting each group with a model so I wanna know wether I should use correlation between those target variable to group them or how should I approach this problem many thanks


I can see three approaches:

1) The one that you mentioned.

2) You may try to form the groups using a clustering technique such as k-means.

3) The selection of the groups is carried by choosing the output variables that share the most relevant variables. You may find the most relevant variables using one of the methods below:

a) Lasso regression (there are today many variations of lasso) and with cross validation to choose the correct regularization parameter.

b) You can run all the subsets of regression considering all variables of interest. However, this is very costly.

c) Partial least squares

d) You may select features considering the importance of the features for out of sample prediction. See the answer to this question

  • $\begingroup$ I think you're talking about selecting the input variables not the output , my problem is that I want to find a way to regroup my output variables so that I can predict each group by one model $\endgroup$ – adam dev Feb 15 at 17:13
  • $\begingroup$ Thanks for the reply btw I really appreciate the effort $\endgroup$ – adam dev Feb 15 at 17:14
  • $\begingroup$ I edited the answer. $\endgroup$ – DanielTheRocketMan Feb 15 at 17:22
  • 1
    $\begingroup$ Thanks a lot , the third approach is pretty interesting didn't think of it that way $\endgroup$ – adam dev Feb 15 at 17:26

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.