# Confidence values for SVM

Hallo community,

similar to my post on GBT I have encountered a similar behaviour with RM's built-in support vector machines with RM 7.6001 (I have not encountered this in earlier versions of RM, but this could just be a coincidence):

Searching the forum, I found a post from @IngoRM stating that "*for binomial classes, a good estimation of the probability for the positive class which is also used by RapidMiner is 1 / (1 + exp(-function_value))) where function_value is the SVM prediction*"

Has this changed, since this statement was made more than 5 years ago? If not, I am a bit confused. I know that confidence values are not the same as probablitites, but why would the algorithm make a (correct) classification of "YES", if the confidence value for this choice is only 0.366?

Any insights or references on this would be very much appreciated. Thank you.

## Answers

3,505RM Data ScientistHi FBT,

that sounds odd. Which SVM are you using?

~Martin

Dortmund, Germany

106UnicornHi Martin,

I am using LibSVM with C-SVC and rbf. This version of SVM worked slightly better with my data, compared to RM's standard SVM.

According to their documentation under 3.2. "SVM Confidence Margin", they seem to calculate the confidence values in a different way then RM does for its native SVM. However, even with their approach I am confused how the results I am seeing can be possible.

Would you know, if RM implements LibSVM and its confidence value calculation exactly as described by its authors?

Thank you.