9.7 Tuning Parameter

• A property of the classifier is that only data points which lie on or violate the margin will affect the hyperplane. These data points are known as support vectors.

• \(C\) controls the bias-variance tradeoff of the classifier.

  • When \(C\) is large: high bias, low variance
  • When \(C\) is small: low bias, high variance

• The property of the SVC solely being dependent on certain observations in classification differs from other classification methods such as LDA (depends on mean of all observations in each class, as well as each class’s covariance matrix using all observations).

• However, logistic regression is more similar to SVC in that it has low sensitivity to observations far from the decision boundary.