There's more…

The main point of the SVM is that a generic problem can always be solved as long as you carefully choose the kernel and all its parameters—for example, going to make a total overfitting of the input dataset. The problem with this method is that it scales quite badly with the size of the dataset, as it is classically attributed to a D2 factor, even if, in this sense, faster implementations can be obtained by optimizing this aspect. The problem is identifying the best kernel and providing it with the best parameters.