For Help on Discriminant function by Genetic Programming!
I want to develope a GP-based disciminant function for a 4-class
problem. Since I try to generate only one discriminant function, it
should map the input vector into different region which means
One of the problem is the definition of boudary region. Since it
is difficult to preset the boundary points,I want to alter these
points using evolution strategy in the process of evolution. And I
also notice that these points should be related to the mimimum ,
maximum , mean value and standart variance of GP mapping value of
each class samples. Do you give me some advice about my idea? Thanks.
The second question is the definition of fitness function. One
way is to use the sum of error percentage of each class or error
percentage of all samples.The shortage is that GP value of samples
are not used. The other way is to use the sum of distance between GP
value of misclassified samples and mean of GP value of this class.
Does anyone tell me some better evlution functions? Thanks.
I am a newer to Genetic Programming. Thank you for your reply.