1
$\begingroup$

I've written Naive Bayesian classifiers before, they work wonderfully. But I'd like a classifier which will learn like a Bayesian classifier and identify new classifications when a new cluster emerges.

Lets presume I have sensor data for some sort of machinery. I have training data for different conditions - too hot, to cold, misfires, etc. Now lets assume some new condition occurs - say a misalignment, or say over revving.

What techniques would allow for that new cluster of data to build a new classification? Naive Bayes+K-Means?

$\endgroup$
5
  • $\begingroup$ Could you elaborate a little more -- naive Bayes is a Bayesian classifier. $\endgroup$
    – suncoolsu
    Commented Aug 15, 2011 at 23:09
  • $\begingroup$ perhaps you want to check the likelihood of the data given the current model, and add a class to the model if the likelihood is low. $\endgroup$ Commented Aug 15, 2011 at 23:20
  • $\begingroup$ @suncoolsu I'm looking for a technique to extend Bayesian classification to be able to learn new classifications not present in initial training sets. $\endgroup$
    – user5847
    Commented Aug 15, 2011 at 23:31
  • $\begingroup$ PLEASE register your account here and on Maths to reclaim your question and be able to edit and post comments. Just go to math.stackexchange.com/login and then to stats.stackexchange.com/login . $\endgroup$
    – user88
    Commented Aug 15, 2011 at 23:35
  • $\begingroup$ @Milton and that's exact what a naive Bayes will do. Basically, you ant to predict (or learn) a new classification based on your previous naive Bayes fit, and choosing to use a Bayesian models takes care of that. (using penalized estimators will give similar answers). $\endgroup$
    – suncoolsu
    Commented Aug 15, 2011 at 23:53

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.