ClassificationSample
Support vector machine
Finds the boundary with the widest margin between the classes, optionally in a kernel-transformed space.
Synthetic data, generated and trained in your browser.
Try this
- 01On Two blobs, lower C: the margin widens and more points become support vectors.
- 02On Moons, switch to RBF: a curved boundary appears.
- 03With RBF, push γ to 300 and turn on the test split: the boundary wraps individual points (overfitting).
How it works
- 1Look for the boundary that separates the classes with the widest margin.
- 2Allow some points inside the margin, at a cost set by C (a soft margin).
- 3Only the points on or inside the margin, the support vectors, define the boundary.
- 4A kernel such as RBF measures similarity in a way that lets the boundary bend.
What the controls teach
- Kernel
- Linear draws a straight boundary; RBF can wrap around clusters.
- C (margin softness)
- Cost of margin violations. Low C gives a wide, tolerant margin; high C fits the training points tightly.
- γ (RBF width)
- Reach of each point in the RBF kernel. High γ makes tight, wiggly boundaries.
Use it when
- Clear margins between classes
- Medium-sized data with complex boundaries (RBF)
Avoid it when
- Very large datasets
- When probabilities are required out of the box