Examples and figures used in this presentation come from:

James, G., Witten D., Hastie, T., Tibshirani, R. "An Introduction to Statistical Learning, with applications in R". Springer, 2013.

You can download it here:
http://www-bcf.usc.edu/~gareth/ISL/ISLR%20First%20Printing.pdf

Predictive models that are calibrated on a set of training data and validated on a set of test data

SVM is typically used as a classification algorithm

• It is an umbrella term for:
  • Maximal margin classifier
  • Support vector classifier
  • Support vector machine

• It is a popular algorithm because:
  • It works well in a variety of settings
  • It sounds cool

• SVM classifies data with a separating hyperplane

• A hyperplane is a flat surface with dimension p-1

• If the data are linearly-seperable, there will be an infinite number of separating hyperplanes

• The goal of the maximal margin classifier is to find a hyperplane with the widest margin between the two classes

• We hope that a large margin on the training data will translate to a large margin on the test data

• Observations on the edge of the margin are called
  support vectors

• If these points were moved slightly, the maximal margin hyperplane would move as well.