The Altair Community is migrating to a new platform to provide a better experience for you. In preparation for the migration, the Altair Community is on read-only mode from October 28 - November 6, 2024. Technical support via cases will continue to work as is. For any urgent requests from Students/Faculty members, please submit the form linked here

How to Analyze the Results of Random Forest Trees

ivonaldosilvaivonaldosilva Member Posts: 4 Learner III
I'm doing a job that involves education data. I need to find a predictor of performance in the midst of data released by the government. So far without problems. I chose the attributes, processed the data, and ran the algorithms. The best answer was the Random Forest model. Well, I ran Random Forest and it generated 20 trees for me. My question is, which tree to use? I need to use one of them, but I have to have a reliable level of security in that choice. can anybody help me? thanks in advance

Best Answer

Answers

  • MartinLiebigMartinLiebig Administrator, Moderator, Employee-RapidMiner, RapidMiner Certified Analyst, RapidMiner Certified Expert, University Professor Posts: 3,533 RM Data Scientist
    keep in mind that the RF uses the average of ALL trees. Thats the trick of a forest, that you use the combined strength of individual random trees.
    I would recommend you to have a look at either Explain Predictions (which is the operator we use to get interpretations in AutoModel and RM GO) or to have a look at the Weights port of the Random Forest.

    Best,
    Martin
    - Sr. Director Data Solutions, Altair RapidMiner -
    Dortmund, Germany
  • ivonaldosilvaivonaldosilva Member Posts: 4 Learner III
    Hi Martin,

    First of all, I'm sorry for the delay in reporting, but I was testing the options in both Random Forest and other options.

    I ran several algorithms to test the model. Naive Bayes, MLM, DT, Perceptron, etc.
    I achieved a very good rating. Accuracy of 96 and error of 4%, Kappa 0.78.

    However, I need to show the results in Árvore, for the sake of my study. If I use the tree created by DT the accuracy is good but the kappa is 0.5. The best alternative for this data set is RF.
    So I ran the RF with 100, 70 and 50 interactions, with depth = 10. So far perfect. But as I have, for example, 50 trees I need to use one of them to illustrate the best result. I could choose any one, but as the study is scientific I have to justify the methodology.

    I also did the Weight of the RF, and perfect I found how much each attribute is relevant to the tree, but I still need the tree illustration.

    Could you give me a tip?

    I appreciate any help

    Ivonaldo
Sign In or Register to comment.