Summary

Maximizing margin classifiers, such as SVM, are a robust alternative to logistic regression for non-linear models for which appropriate kernel functions exists. Moreover, SVM is less demanding of computation resources for very large datasets.

In a nutshell, this chapter introduces you to the basic concept of kernel functions and the theory and application of SVM classifiers as applied to financial instruments. The chapter concludes with the one-class SVM classification for detecting outliers and an overview of the support vector regression models.

As with other discriminative models, the selection of the optimization method for SVMs has a critical impact not only on the quality of the model, but also on the performance (time complexity) of the training and cross-validation process.

This chapter concludes our overview of discriminative, supervised machine learning models. The next couple of chapters deal with a new universe: evolutionary models and reinforcement learning.