Reputation: 21
I'm using the MultiClassSupportVectorMachine class to do some classification. Specifically, my data has 24 dimensions, with the values being grouped pretty close together. I will be identifying around 10 or so classes in this data.
I'm looking to identify when an inputted value is really far off from the groups. Something along the lines of having a class 0 which would be unidentified and then having classes 1 to 10 only being outputted when the SVM has a high degree of confidence it is in the group.
In essence I am looking to go from the top to the bottom of this image: this image showing SVM classification
Is something like this possible in accord.net?
Thanks!
Upvotes: 1
Views: 106
Reputation: 21
I answered my own question!
This can be accomplished by using the Probability function to get an estimate of how accurate the guess is from the SupportVectorMachine class, and then using that as a threshold to reject guesses with a low probability.
Upvotes: 1