How to run regression with qualitative predictor variable

[yuxiliu]

Hi! I am working on finding the best predictor variables for a dependent variable. I have been using the Linear Regression tool to find the correlation with the quantiative variables. But Linear Regression only allows numeric/binary values.  What tool/s do you recommend for finding correlation with the qualitative variables (i.e. categories) ?

Thanks so much!

[JohnJPS]

Just to make sure I understand: you're setting one of your categorical variables as the target of your Linear Regression, and looking for correlation with other variables that way?

If so, that's really a classification problem rather than a regression problem: you want the system to predict (classify) the category value of the categorical variable. Therefore, any of the classification-type model algorithms should work better... (Boosted Model, Decision Tree, Decision Forest, Naive Bayes, SVM, etc...)

Hope that helps!

[yuxiliu]

Thank you JohnJPS. 

I am setting my categorical variable as independent and a quantitative variable as dependent. I think your recommendation works for this setup. Any thoughts on which of the classification-type model is best for relating Sales data (dependent) to customer categories (independent)? 

[JohnJPS · Accepted solution]

Unfortunately, I wouldn't be the one to hazard a guess as to when one classifier might be better than another.  No guarantees, but the Boosted Model is hard to beat in most cases, so would probably be a safe bet.  (The implementation in Alteryx is based on R's "gbm" which you could google for more information if desired.)

[RodL]

I believe that help in answering @yuxiliu's question of which classifier is better would come from the relatively new Model Comparison tool.

You can find a sample of this with the macro on the Gallery in the Predictive District.

[yuxiliu]

Awesome! Thanks for the recommendation!