Support Vector Machines

SVM finds the hyperplane that maximizes the margin between classes. Current margin: 0.49 — try different angles to find the widest street.

Only points on the margin boundary determine the decision surface — these are support vectors. The dragged point is far from the margin — moving it has no effect on the boundary.

Moderate C (1.0): Balanced trade-off between margin width and classification errors.

Polynomial kernel: implicitly maps data to higher dimensions where curved boundaries can separate non-linear patterns.

Moderate γ (0.40): Balanced boundary flexibility. Good generalization between training and test data.

For 3 classes, SVM trains 3 one-vs-rest classifiers and picks the highest confidence. Each of the 3 classifiers learns to separate one class from all others.