Unfortunately for us, most data is not linearly separable.
For example, look at the data on the right. It is clearly not linearly separable.
For a simpler classification algorithm, that might be it and we would have to find another way to classify the data, but not SVMs!
To find a classifier for non-linearly separable data using SVMs, we have to do a few things. Click through the steps to see:
First, we start out with out dataset that can't be classified using a linear classifier.
The method outlined above is very useful, but depending on the number of dimensions, the complexity can grow extremely quickly until it becomes too difficult to calculate. To counter this, Support Vector Machines have another trick, this one known as the Kernel Trick.