Thread Subject:

Bayesian Classification

Can anyone please guide me to understand how matlab performs classifcation using bayesian classifier in version 2011(a).
 For clarity if we have a large data set with two classes and a new data is to be tested to belong to any one of the classes . How does classify performs this classification using diaglinear, as the conditional probability density function is not known and also if we are not allowed to make any assumptions regarding the distribution of the data... such as gaussian.

So whats the logic behind the classification performed.

"Biman Chandra Dey" <deybimanchandra@yahoo.com> wrote in message <jsbbdn$r98$1@newscl01ah.mathworks.com>...
> Can anyone please guide me to understand how matlab performs classifcation using bayesian classifier in version 2011(a).
> For clarity if we have a large data set with two classes and a new data is to be tested >to belong to any one of the classes . How does classify performs this classification using >diaglinear, as the conditional probability density function is not known
> and also if we are not allowed to make any assumptions regarding the distribution of >the data... such as gaussian

No. Reread the documentation

help classify
doc classify

Distributions are assumed to be Gaussian with identical diagonal covariance matrices.
 
> So whats the logic behind the classification performed.

Maximum estimated posterior probability

Hope this helps.

Greg

"Greg Heath" <heath@alumni.brown.edu> wrote in message <jsd2lq$nrd$1@newscl01ah.mathworks.com>...
> "Biman Chandra Dey" <deybimanchandra@yahoo.com> wrote in message <jsbbdn$r98$1@newscl01ah.mathworks.com>...
> > Can anyone please guide me to understand how matlab performs classifcation using bayesian classifier in version 2011(a).
> > For clarity if we have a large data set with two classes and a new data is to be tested >to belong to any one of the classes . How does classify performs this classification using >diaglinear, as the conditional probability density function is not known
> > and also if we are not allowed to make any assumptions regarding the distribution of >the data... such as gaussian
>
> No. Reread the documentation
>
> help classify
> doc classify
>
> Distributions are assumed to be Gaussian with identical diagonal covariance matrices.
>
> > So whats the logic behind the classification performed.
>
> Maximum estimated posterior probability
>
> Hope this helps.
>
> Greg

Thank you sir for the explanation but being new to matlab i still have some daubts...
 let me make my problem more clear..

I have 1000 colour images from that i made a datbase of the GREEN component of the pixels
and separated them into classes SKIN and WOUND

Now if i need to perform pixel wise classification of a new image into the
classes of wound and skin using Bayes classifer, by using the database created. How can i do that?
I am trying to implement it in matlab and dont want to use the inbuilt command
CLASSIFY as i have all the data and dnt want to assume anything about the distribution.

"Biman Chandra Dey" <deybimanchandra@yahoo.com> wrote in message <jse77s$juf$1@newscl01ah.mathworks.com>...
> "Greg Heath" <heath@alumni.brown.edu> wrote in message <jsd2lq$nrd$1@newscl01ah.mathworks.com>...
> > "Biman Chandra Dey" <deybimanchandra@yahoo.com> wrote in message <jsbbdn$r98$1@newscl01ah.mathworks.com>...
> > > Can anyone please guide me to understand how matlab performs classifcation using bayesian classifier in version 2011(a).
> > > For clarity if we have a large data set with two classes and a new data is to be tested >to belong to any one of the classes . How does classify performs this classification using >diaglinear, as the conditional probability density function is not known
> > > and also if we are not allowed to make any assumptions regarding the distribution of >the data... such as gaussian
> >
> > No. Reread the documentation
> >
> > help classify
> > doc classify
> >
> > Distributions are assumed to be Gaussian with identical diagonal covariance matrices.
> >
> > > So whats the logic behind the classification performed.
> >
> > Maximum estimated posterior probability
> >
> > Hope this helps.
> >
> > Greg
>
> Thank you sir for the explanation but being new to matlab i still have some daubts...
> let me make my problem more clear..
>
> I have 1000 colour images from that i made a datbase of the GREEN component of the pixels
> and separated them into classes SKIN and WOUND
>
> Now if i need to perform pixel wise classification of a new image into the
> classes of wound and skin using Bayes classifer, by using the database created. How can i do that?
> I am trying to implement it in matlab and dont want to use the inbuilt command
> CLASSIFY as i have all the data and dnt want to assume anything about the distribution.

I recommend using CLASSIFY first.

Hope this helps.

Greg

"Biman Chandra Dey" <deybimanchandra@yahoo.com> wrote in message <jsbbdn$r98$1@newscl01ah.mathworks.com>...
> Can anyone please guide me to understand how matlab performs classifcation using bayesian classifier in version 2011(a).
> For clarity if we have a large data set with two classes and a new data is to be tested to belong to any one of the classes . How does classify performs this classification using diaglinear, as the conditional probability density function is not known and also if we are not allowed to make any assumptions regarding the distribution of the data... such as gaussian.
>
> So whats the logic behind the classification performed.


the classify command in matlab assumes the distribution of the given data to be gaussian, then estimates the covariance and based on it finally assigns a class to the test data.... i have used classify but the results were not very encouraging resulting to a lot of missclassification... I would request if there are some other ways to perform the classification ... such as we can use the histogram of the data selected , normalise it and the find its pdf.... but i am a bit confused on implementing it...

I would be obliged to obtain some help on my daubt..
thanks..