Skip to main content

Support Vector Machines

  • Support vectors are margins added around the decision boundary to allow the model to feel the data
  • Soft margin
    • allows for instances to be within the margins
    • most common type of SVM
  • Hard margin
    • no instances are allowed to be within the margins
  • The width of the margins are typically denoted with Ïĩ\epsilon
  • The construction of the error function is to ensure flatness by minimizing the coefficients to ensure there are no overfitting and minimize the residual error to be less than the margin width Ïĩ\epsilon
    • The model may not exist to satisfy the hard margin condition, leading to a surrogate function using slack variables which is the soft margin condition
    • Solving this problem involves [[Quadratic Programming]]