e support vector machine algorithm

vector machine (SVM) was proposed by Vladimir Vapnik [Cortes

nik, 1995]. The basic principle of SVM is to use a kernel function

raw data space to a kernel space, where classification analysis or

n analysis can be implemented [Aizerman, et al., 1964; Cortes

nik, 1995; Bishop, 2006]. It is also hypothesised that such a

space is linear.

FNN, all radial basis functions will have their weights estimated

ero values. In reality, some radial basis functions may be

t. In addition, the maximised likelihood approach [Duda, et al.,

hop, 2006; Rossi, 2018] used to estimate model parameters using

squared errors approach may not be able to generate an optimal

searching for the best hyperplane for the discrimination between

es of data points, SVM finds a set of data points which are most

o classify. These data points are referred to as the support vectors.

the closest to the hyperplane and are located on the boundaries

ssification margin between two classes.

dea proposed in SVM is to optimise the classification margin in a

space. Figure 3.29 shows this principle. In Figure 3.29(a), a wider

tion margin will allow some data points which can be easily

to fall in. In Figure 3.29(b), no such data point is found within

ification margin. SVM will optimise the classification margin

he search of the support vectors. No data points other than support

will be used as kernels, hence the model is more parsimonious than

N model.

M classifier is more robust to data noise or outliers due to the

ion of a classification margin. The generalisation capability is

boosted. A trained SVM classifier is a linear combination of the

es between an input and support vectors. The similarity between

vector and a support vector can be quantified by a kernel function,

defined as below, where ܠ௡ is the n^th support vector,