fitcnb, Naive Bayes classifier predictor distributions

Dear All,

I plan training a naive Bayes classifier on using two features predictors (assumed independent, and both derived from voxel intensities in an image) for a binary (0,1) voxel/pixcel classification task (within-label, outside-label). The two predictors have a generalized extreme value distribution (GEVD), so they do not fit the predictor distributions includied with fitcnb. What distribution should I adopt, from those included?

Thank you.

OL

[Purvaja]

Hey @Octavian,

You can use kernel density estimation in fitcnb for automatic non-parametric modeling of GEVD features, or implement manual Naive Bayes using gevpdf to compute class-conditional likelihoods.