How to select the optimal/best tree from the collection of trees generated by Random Forest?

hsanchezhsanchez Member Posts: 10 Contributor II
I have a process that utilizes three random forest algorithms each using different parameters. I use the operator collect and compare models to select the best of the three learners. However, I would like to know which of the collection of trees was chosen by the random forest learner; hence I can use that tree to explain my finding. I know I can use simulation operator, prescriptive analytics, but I need the best tree from the collection of the random forest best learner. Can anyone give me a clue? Thank you
Tagged:

Best Answer

Answers

  • mschmitzmschmitz Administrator, Moderator, Employee, RapidMiner Certified Analyst, RapidMiner Certified Expert, University Professor Posts: 3,231 RM Data Scientist

    a random forest uses the average prediction of all these trees. Thats the trick of a RF.

    Best,
    Martin
    - Head of Data Science Services at RapidMiner -
    Dortmund, Germany
    Tghadially
  • hsanchezhsanchez Member Posts: 10 Contributor II
    yeah...I kind of know that but it is hard to explain the findings...I was expecting that I could pick the best of all; or the software may recommend me based on certain criteria which of those trees perform the best. Thank @mschmitz.
  • mschmitzmschmitz Administrator, Moderator, Employee, RapidMiner Certified Analyst, RapidMiner Certified Expert, University Professor Posts: 3,231 RM Data Scientist
    Hi,
    well, but then you essentially do a decision tree (or better random tree). The problem with these are that the predictive accuracy of them is worse. You are in the 'classic' problem of Interpretabilty vs predictive accuracy. I would recommend to use Explain Predictions and the model simulator. Maybe RF encoder can be a help here. It gives you the individual predition of each tree, but this is also tricky.

    Best,
    Martin
    - Head of Data Science Services at RapidMiner -
    Dortmund, Germany
    Tghadially
Sign In or Register to comment.