Maximize AUC

pb42
pb42 New Altair Community Member
edited November 5 in Community Q&A
I recognize that this is a general question, but I am seeking some general guidance on approaches to consider to improve the AUC of a model. The data set I have is student data, and I need to develop a model that will predict the outcome of a student (successful/not successful). The results will be assessed by the AUC (yes, this is a course assignment). I have tried many models and many ensemble models, but am still falling short of the AUC that is set by the professor.
Tagged:

Best Answers

  • BalazsBarany
    BalazsBarany New Altair Community Member
    Answer ✓
    Hi!

    There are many ways to get better models (which will lead to a higher AUC).

    Some approaches:
    - Attribute selection (remove irrelevant or noisy attributes)
    - Attribute generation (sometimes the relation between attributes, their product or division or difference are also relevant indicators that help the model make better decisions)
    - Downsampling the majority class if you have imbalances in the data, or weighting the minority class
    - Model parameter optimization

    Search for these techniques in the Academy, there are videos for all of them.

    Regards,
    Balázs

Answers

  • BalazsBarany
    BalazsBarany New Altair Community Member
    Answer ✓
    Hi!

    There are many ways to get better models (which will lead to a higher AUC).

    Some approaches:
    - Attribute selection (remove irrelevant or noisy attributes)
    - Attribute generation (sometimes the relation between attributes, their product or division or difference are also relevant indicators that help the model make better decisions)
    - Downsampling the majority class if you have imbalances in the data, or weighting the minority class
    - Model parameter optimization

    Search for these techniques in the Academy, there are videos for all of them.

    Regards,
    Balázs