Looks for a line that separates classes

Inherently do binary classification, but may be extended

Line that maximize distance (MARGIN) to nearest point : robustness of result

First correct classification then maximize margin

Ability to ignore outliers

Decision Boundary : Line or a hyperplane. To generalize the simple idea of finding the linear separation to non-linear boundary the solution is to move to a different feature space where features are linearly separable than back to original features) using the kernel parameter. For example, by adding a feature z ($ z = x^2 + y^2 $), SVM is able to find non linear separations such as center-surround separation.

Parameters : kernel, C, gamma

• C : controls tradeoff between smooth decision boundary (low C) and classifying training points correctly (high C)
• Gamma : defines how far the influence of a single training example reaches the boundary. (low values -far, high values - close)
• kernel: tranforms inseparable features to separable features

• https://classroom.udacity.com/courses/ud120/