I am working in a 3D gesture recognition project with kinect. So far, I have trained classifiers with 5 gestures. Still need to implement for real time recognition. The thing is this. Most of the times the continuous flow of data will be "no gesture". However, my classifier will match with any of the classes even though the likelihoods would be minimum.

The number of no gestures options is practically infinite so acquiring data from random movements as a additional class becomes impractical.

I want to know if Sklearn libraries give any method for setting some thresholds in order to reject outliers.

Basically something as explined here, but in python. http://www.nickgillian.com/wiki/pmwiki.php/GRT/AutomaticGestureSpotting

  • 2
    $\begingroup$ Easy, use the decision values instead of predicted labels and put some threshold on them. For example, for probabilistic classifiers the default threshold is a probability of 0.5, but nothing stops you to increase it to 0.8 to obtain a more conservative model (i.e., one with higher recall and lower precision). $\endgroup$ – Marc Claesen Sep 13 '15 at 10:16

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.