performance cost

ThiruThiru Member Posts: 100 Guru
edited February 2020 in Help
hi all,

1. when I use classifier performance ( cost) operator, im not able to get confusion matrix in the results.  Is there anyway to get both cost as well as conf matrix.

2. Im getting cost as 1.774.   what does it mean really.  can we derive this value in terms of currency like dollars/rupees.

3.  Also would like to know use of meta cost operator and how does it 
differ from normal cost operator


regds

thiru
Tagged:

Answers

  • MartinLiebigMartinLiebig Administrator, Moderator, Employee, RapidMiner Certified Analyst, RapidMiner Certified Expert, University Professor Posts: 3,503 RM Data Scientist
    Hi @Thiru,
    you can use both, Performance (Classification) and Performance (Cost). The former one has a input for another Performance vector to create one big one.

    2. The average cost per example is 1.774. I recommend to compare this to a cost of a default model (Default model operator), which could be you predict all to be Loyal or something.

    3. Check: https://dl.acm.org/doi/10.1145/312129.312220 which is the paper for it. It's not too crazy. You use a bagged ensemble to get a better estimate of confidence. This is actually not much like the Performance cost, but more like the Cost Sensitive Scoring operator we use in AutoModel.

    Cheers,
    Martin
    - Sr. Director Data Solutions, Altair RapidMiner -
    Dortmund, Germany
  • ThiruThiru Member Posts: 100 Guru
    hi @ martin - thanks for your suggestions and it works.  I just wanted to further calrify,

    a.   what I understand, the default is the base reference for comparisons with new faults and work out the misclassification costs. first of all  how this default model is chosen ?   


    b.  How to compare two or more classifiers - in a single process flow. 
     ( I would like to compare two or more classifiers to arrive  the misclassification cost in the single process)"                                                                                                      
    thanks
    thiru                                              
Sign In or Register to comment.