SVM

What is it

• method for supervised classification, regression or outlier detection
• devides the classes by a gab (margin) as wide as possible (https://en.wikipedia.org/wiki/Support_vector_machine)
  • by giving a vector that describes the plane to divide the classes
  • if data is not sepbarable, one can add a slack variable to allow errors
• linear and non-linear with kernel trick
  • rbf is most common kernel

pros:

• effective in high dimensional space
• kernels can be specified for decision functions

cons:

• SVMs do not directly provice probability (maybe use Platt scaling for this?)

Scikit has overview of support vector machine functions: http://scikit-learn.org/stable/modules/svm.html

Mathematical background: http://www.cs.columbia.edu/~kathy/cs4701/documents/jason_svm_tutorial.pdf