SVM diagrams (SVM 2D)

2D SVM interpretation

A lot of diagrams show hyperplane with parallel lines, but that could be confusing, because you are searching for widest gap between two classes with support vectors assisting in finding it. Usually data is not 2D (higher dimensional data could be reduced to 2D), so the separating plane in higher dimensions could be even more confusing to imagine if you think of parallel lines and that you may need to only use 3-4 support vectors, in reality you can use many and come up with hyperplane which is hypercurved surface, which if you transfer in to different dimension can be flat. In the mathematical version of SVM (non visual) hyperplane starts as line equation, then is gradually transformed to quadratic optimization formula, which needs variables called alphas to be found, then it can be used to enter new vector, run through new formula and tell if the result is -1, or 1 indicating the class label. More text based explanations will be in further posts.

Figure 1. SVM and hyperplane separation (Source: Ubaby, 2016)