ROC and LIFT charts

theotheo Member Posts: 2 Contributor I
edited November 2018 in Help
Hello everyone

I'm an italian student and I approached to Rapid Miner just recently.

My professor gave me a dataset to analyze which consist of a training set (15.309 record) and a test set (15.308 record). The data types are: nominal, integer, real and numerical.
In order to achieve a good score, I should compare 2-3 different types of CLASSIFICATIONS model (I choose Naive Bayes and Logistic Regression). Then plot the results into a ROC chart and a LIFT chart.

The problem is that I'm unable to plot the confrontation (I cannot find the correct node to use).
I guess I have to use the LIFT CHART and COMPARE ROCS nodes but they won't work.
Even if the process does not contain "red nodes" it seems that there's an error into the original dataset or inside some node; still I can't find out where the fault is.

Thanks a lot

PS.
I'm running a process just right now and it is taking about 1 hour to complete. Is that right?

Answers

  • landland RapidMiner Certified Analyst, RapidMiner Certified Expert, Member Posts: 2,531 Unicorn
    Hi,
    the logistic regression takes some time. I would replace it by an SVM, that should be somehow faster in many cases. Since you don't gave any details, not even the error message, I don't think that you want to ask for any help, are you? If you do, you should at least give us the detailed error message AND post the process here inside a code area. You can simply copy it from the XML tab.

    Anyway I would suggest to first take a look at the sample processes that come together with RapidMiner. There are some showing how ROC and LIF CHART generation works.

    Another hint: On our website are videos available showing how to use RapidMiner, could be of benefit for you.

    Greetings,
     Sebastian
  • theotheo Member Posts: 2 Contributor I
    Hello

    i would like to thank You for Your answer. I solved the problem by looking the samples process and simply copying some nodes (as You suggested).

    Many Thanks and good day to You
Sign In or Register to comment.