Introduction to Pattern Recognition: A Matlab Approach

2.5 SVM: The Nonlinear Case 55

Table 2.5 Results for the SVM Classifiers Designed in Steps 2,

3, and 4 of Example 2.5.1

Training Error Testing Error No . Support Vectors

Linear 7.33% 7.33% 26

RBF

(0.1) 0.00% 32.67% 150

RBF

(2) 1.33% 3.33% 30

poly

(5,0) — — —

poly

(3,1) 0.00% 2.67% 8

Note: RBF(a) denotes the SVM classifiercorresponding to the radial basis kernel

function with σ = a;poly(n, β) denotes the SVM classifer with the polynomial

kernel function of the form (x

T

y +β)

n

. The algorithm does not converge for the

case poly(5,0).

Step 4. To generate a nonlinear SVM classifier with the polynomial kernel function using n =3and

β =1, work as before but now set

kernel='pol ...