I want to evaluate my model with help of a confusion matrix. Which measure can I use to consider that my data is interval scaled (eg.:1,2,3,4,5,6,7,8,9,10) ? As I understand the Kappa measure is often used for nominal scaled data (eg.: red, green blue). If I use the Kappa estimate for the interval scaled data a misclassification of 1 --> 2 (1) is considered to be as severe as a misclassification of 1 --> 8 (2). However, the misclassification (1) is much less sever than the misclassification (2). How can I account for this in a confusion matrix?

  • 1
    $\begingroup$ you can use weighted kappa, if you use r, try reading the help from kappa2. $\endgroup$ – ceoec May 13 '15 at 8:07
  • $\begingroup$ In kappa 2 I should use the '"squared" weight for a interval scale for the above mentioned example, right? $\endgroup$ – kalakaru May 13 '15 at 11:18
  • $\begingroup$ yes i believe so... $\endgroup$ – ceoec May 13 '15 at 13:07
  • $\begingroup$ are you also familiar with the cohen.test from the psych package? If yes, can you maybe answer this question: stats.stackexchange.com/questions/152799/… $\endgroup$ – kalakaru May 18 '15 at 9:27

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.