Predict an unbalanced class

mfulgerimfulgeri Member Posts: 3 Contributor I
edited December 2018 in Help

Hi guys, 

I'm pretty new with RM so apologize if I'm asking someting straightfoward.

I'm trying to train a predictor on a class that is heavely unbalanced - say churn rate(on average I have 1 observation with chur every 999 observation without).

I read that boosted trees could be a good choice as algorthm, but I have some difficulties to understand if there is a way to have a "penalized" version of this.. how can I achieve this? 

Another approach that I could follow would be to generate new data based on the distribution of the other variables in the data set, is that possible to do in RM?

 

many thanks for your help! 

Matteo 

Tagged:

Answers

  • sgenzersgenzer Administrator, Moderator, Employee, RapidMiner Certified Analyst, Community Manager, Member, University Professor, PM Moderator Posts: 2,959 Community Manager

    Hello @mfulgeri - welcome to the community.  Good question.  So I assume first of all that you've seen this video on churn prevention?  Second I would investigate the use of the Sample operator, using the "balance data" checkbox advanced parameter:

     

    Screen Shot 2017-11-15 at 3.31.03 PM.png

     

    [EDIT: forgot to mention that I always use mod.rapidminer.com as a resource when trying to decide which learner would be most appropriate.  It's a very under-utlilized and amazing resource.]

     

    Scott

  • Telcontar120Telcontar120 Moderator, RapidMiner Certified Analyst, RapidMiner Certified Expert, Member Posts: 1,635 Unicorn
    There was an almost identical question asked a few months ago, and rather than repeating my earlier somewhat detailed answer, I encourage you to check it out.
    https://community.rapidminer.com/t5/RapidMiner-Studio-Forum/Dealing-with-imbalanced-data/m-p/41385
    In short there are many options for dealing with class imbalance in RapidMiner.
    Brian T.
    Lindon Ventures 
    Data Science Consulting from Certified RapidMiner Experts
  • sgenzersgenzer Administrator, Moderator, Employee, RapidMiner Certified Analyst, Community Manager, Member, University Professor, PM Moderator Posts: 2,959 Community Manager

    thank you @Telcontar120 - I was looking for that post but could not find it!


    Scott

     

Sign In or Register to comment.