Calculations for the Pos and Neg Predictive Values and the PSEP in binomial performance
Hi. I am trying to understand a few calculations in the Binomial Performance Classification operator. I am using the Titanic dataset, and a decision tree operator within the nominal cross-validation building block. I switched the performance operator for the binomial classification performance operator so that I could get more criterion. Everything looks great (meaning I can verify the values provided), except for three values.
The value Rapidminer shows for the Positive Predictive Value is 82.87%, which matches my calculation for a Negative Predictive Value. The reverse is also true that the Negative Predictive Value of 75.49% matches my Positive value. Is there a labeling mis-match? There are more negative (no) survival values in the Titanic dataset than positive (yes) values, so I think my values are correct.
Also, how do you calculate the PSEP or Positive Satisfactory Error Probability value of .584? The equation I use is FPR + FNR * (1- Acceptable Error Rate), but after substituting FPR and FNR values the only value that matches your score for the Acceptable Error Rate is -0.532. But it doesn't make sense that an Acceptable Error Rate is a negative value, nor do I understand how you arrive at 53.2%?
Can someone help explain these differences to me?
Thanks for you time.