Options

"How to convert confidence attribute to probabilty?"

komeil_shaerikomeil_shaeri Member Posts: 13 Contributor II
edited June 2019 in Help

Hi, 

I wondering how can I get probabilty of having Yes or No instead of confidence attribute in classification models. In other words, I need to get the probability of each instance (example) becoming Yes or No. I have tried "Generate Attribute" but I couln't get it. 

 

Thanks, 

Komeil

Answers

  • Options
    MartinLiebigMartinLiebig Administrator, Moderator, Employee, RapidMiner Certified Analyst, RapidMiner Certified Expert, University Professor Posts: 3,524 RM Data Scientist

    Hi,

     

    the confidence can be interpreted as a propability. It's a value between 0-1 which sums up to one for all classes. So what do you want to have otherwise?


    ~Martin

    - Sr. Director Data Solutions, Altair RapidMiner -
    Dortmund, Germany
  • Options
    komeil_shaerikomeil_shaeri Member Posts: 13 Contributor II

    Do all the instances have the same probability of 0% and 100%? Of course not. Some of them are marginally classified into a group but the others are becoming part other groups with very high probabilities. There should be way to find the exact  probabilities for each instance ...  

    Thanks,

  • Options
    MartinLiebigMartinLiebig Administrator, Moderator, Employee, RapidMiner Certified Analyst, RapidMiner Certified Expert, University Professor Posts: 3,524 RM Data Scientist

     

    Sorry, but i do not get your point.

    They get confidence according to the properties. The confidence is as exact as a double can be. Do you confuse prediction and confidence?

     

    ~Martin

    - Sr. Director Data Solutions, Altair RapidMiner -
    Dortmund, Germany
  • Options
    komeil_shaerikomeil_shaeri Member Posts: 13 Contributor II

    The problem is solved. It was interesting because in one data set, confidence attributes are just 1.00 and 0.00 but in another data set there are lots of variations in the numbers (e.g. 0.755 and 0.245). 

     

    Thanks,

Sign In or Register to comment.