Support Vector Machine Flashcards
Support Vector Machine
To establish the broadest possible “street” between distinct classes.
Support vector
Any instance located on the “street”.
Decision boundary
Entirely determined by the support vectors
2 Type of Support Vector Machine
SVM Classification - Fit the largest possible street between two classes while limiting margin violation
SVM Regression - Fit as many instances as possible on the street while limiting margin violations
2 Type of Support Vector Machine Margins (Only for classification)
Hard Margin Classification
Soft Margin Classification - To find a balance between maintaining the street as wide as possible while also limiting the number of margin violations.
2 Characteristic of Hard Margin Classification
All instances must be off the street
All instances must be on the right side
2 Issue of Hard Margin Classification
Works only when data is linearly separable.
Sensitive to outliers.
2 Approach of Nonlinear SVM Classification
Approach 1 - High polynomial degree
Approach 2 - Polynomial Kernel (kernel trick) - Higher dimensional; Not add new features
C (Regularization Hyperparameter)
Controls the trade-off between achieving a large margin and minimising classification errors
Small value of C - Wider margin & some misclassifications (margin violations)
Large value of C - Narrower margin & potentially overfit