Maximum margin, soft margin and kernel methods for classification, with Python examples and finance use cases
|
Support Vector Machines (pdf)
|
3.88 M |
In a support vector machine, two classes are separated by the hyperplane with the widest margin, and the hyperplane is defined by the training points closest to it, the support vectors. The linear case is derived as a constrained optimisation problem and solved in its dual form. Some misclassified points are tolerated with slack variables and a penalty parameter C, and non-linear boundaries are obtained with the kernel trick, without an explicit feature map, using polynomial, radial basis function and sigmoid kernels. In Python, linear and kernel classifiers are fitted to a moons data set and in-sample accuracy is compared with out-of-sample accuracy. Finance applications are credit scoring, fraud detection, loan default probability, yield curve prediction, derivative pricing and a Bitcoin trading strategy.