Introduction to Support Vector Machine

Decision Bound

A decision bound is a boundary chosen by the classifier to determine if data points are within a certain class or another. Typically a decision bound is a function \(f: x \in \mathbb R^d \rightarrow \mathbb R\) such that \(f(x)=0\).

Hyperplane as Decision Bounds

A hyperplane is a \(p-1\) dimension subspace living in \(p\)-dimension. The hyperplane is mathmetically described as the equation,

\[ \theta \cdot x + \theta_0 = 0 \]

\[ \theta_0 + \theta_1x_1 + \theta_2x_2 + \ldots + \theta_px_p = 0 \]

• \(\theta\) : A vector normal to the hyperplane. This determines the direction that the hyperplane points to.

• \(\theta_0\) : The intersect point which may be interpeted as the distance the hyperplane is from the origin.

We may use the hyperplane as a decision bound. In 2D, the hyperplane is a line. In this case the decision function is thought of as:

If the input falls on the left of the line then it’s class A otherwise class B.

In math, the decision function takes in an input vector \(X_i\) and outputs a label (strictly this is numerically represented),

\[\begin{split} y = \begin{cases} \text{1}, & \theta \cdot X_i > 0\\ \text{-1}, & \theta \cdot X_i < 0 \end{cases} \end{split}\]

• \(y=1\) can be represent class A and \(y=-1\) can be class B

• You can immediately tell that hyerplane are great for linearly separable data