(a) (b)

(a) A classification margin, which may not be optimal. (b) A classification

ich may be optimised. The dashed line in each panel is a classifier, a hyperplane.

ines in each panel stand for the classification margin or boundaries. The open

triangles stand for easily classified data points. The filled circles and triangles

upport vectors which are exactly on the classification boundaries. The shaded

triangles are those data points which cannot be correctly classified. The dotted

triangles are those fall in the classification margin.  is the width of the

on margin.

ose ݕ௡is the class label of the n^th support vector and ߙ௡ is the

parameter of the n^th support vector and the number of support

K) is determined by an SVM algorithm for a specific data set. A

del makes a decision using the following formula,

ݕොൌsign ቌ෍ߙ௡ݕ௡

௄

௡ୀଵ

ߖሺܠ, ܠ௡ሻቍ

(3.67)

M, ߖሺܠ௡ሻ is referred to as a feature and is defined as below,

ߖሺܠ, ܠ௡ሻൌߖሺܠሻ⋅ߖሺܠ௡ሻ

(3.68)

e 3.30 shows a SVM model constructed for a simulated data set,

can be seen that the support vectors are the data points which are

n the boundaries of two clusters.