Key concepts for SVC
We can use support vector machines (SVMs) to find a line or curve to separate instances by class. When classes can be discriminated by a line, they are said to be linearly separable.
There may, however, be many possible linear classifiers, as we can see in Figure 13.1. Each line successfully discriminates between the two classes, represented by dots and squares, using the two features x1 and x2. The key difference is in how the lines would classify new instances, represented by the transparent rectangle. Using the line closest to the squares would cause the transparent rectanglez to be classified as a dot. Using either of the other two lines would classify it as a square.
Figure 13.1 – Three possible linear classifiers
When a linear discriminant is very close to training instances, as is the case with two of the lines in Figure 13.2, there is a greater risk of misclassifying new instances. We want a line that gives us the maximum...
