baseball regression model

[shahd]

Hi all
I have a baseball data sets with 30 independent variables. In this data set, I have one variable which is a combination of the summation 3 variables from the data set. For example, x8=x3+x4+x5. I need to build a multiple linear regression model, if i include x8 in my model should i remove x3,x4,x5. Could you please advise with this

[Reeza]

Yes you should.

[shahd]

What is the best choice to remove x8 or x3, x4,x5

[Reeza · Accepted solution]

Depends on the data, I'd run both models and pick the one that does better. Consider x8 an engineered feature.

[shahd]

Thank you very much. This is very helpful. what if I have multiplication not addition should I follow the same procedure

[Reeza]

I would probably treat that as two variables and include an interaction term instead. Then if the interaction term isn't significant you would drop it. It's a non-linear relationship instead of linear in your first question. However, theoretically if you're model building you should be testing these things at the same time ie you have two different options for two sets of variables, so that means fitting 4 models and determine what's the best. Or three if you take my advice. Whatever you do, document it, and explain your decision making in your notes.

[shahd]

Thanks a lot.
I have another variable x9=(x1+x7+x8)/(x3+x4+x6) could you please advice with this

[Reeza]

Now I think you need to read up on feature engineering.

[Reeza]

Also, don’t know how much of this is made up, but you’ve mentioned three different features and they all overlap with calculations so you’ll need to factor that in as well. Basically if those were your actual variables and situation it’s more complex than originally you suggested.