Naive Bayes Classification - Chanig distribution from Normal/Gaussian to other types

I am using Naive Bayes for multi-classification using:

Mdl = fitcnb(Features,Label);

Naive Bayes uses a normal distribution but when I checked my data, it doesn't fit as shown in the attachments.

Is there a way to change the type of distribution that best fits my data?

Thanks in advance.

Answer 1

[Hrishikesh Borate]

Hi,

It’s my understanding that you are using Naïve Bayes classification for a multiclass problem with fitcnb function, and want to try some other distribution apart from Normal Distribution.

You can use Multinomial Distribution, Multivariate Multinomial Distribution, Normal Distribution. To use these, you must specify them as

Mdl = fitcnb(___,Name,Value)  ;

where,

Name will be ‘DistributionNames’, and

Value can be ‘mn’ (for Multinomial Distribution), ‘mvmn’ (for Multivariate Multinomial Distribution), ‘normal’ (for Normal Distribution).

For more information, refer to fitcnb Documentation

[Oliver Ramos]

Thank you very much Hrishikesh!

Are you familiar also with discriminant analysis (fitcdiscr)? Are there also other 'DistributionNames' like mvmn, kernel, and mn, for fitcdiscr? I'm thinking of using fitcdiscr to remove the assumption of conditional independence.