SVM Model with Unequal Sample Number Between Classes
Jian Xiao · 2003
The paper briefly introduces the principle of SVM in sorting first. Then using normal distribution as an example, the paper points out that the SVM can not get excellent sorting ability when the sample numbers of different class are different greatly. A method has been designed to compute the penalty parameter to error samples. On the basis of different penalty parameter an advanced SVM algorithm has been developed. Two simulated examples show the new algorithm is feasible.