Description
Support Vector Machines find the maximum-margin hyperplane (linear or kernel-induced) separating classes. Kernel SVMs were the dominant non-DL classifier for CSI sensing pre-2017 and remain a competitive baseline on small datasets, particularly when paired with carefully engineered subcarrier features.
When it's used
- Small-dataset CSI classification baselines
- Hand-crafted-feature pipelines on embedded hardware
- One-vs-rest open-set detection prototypes
Limitations
- Scales poorly above ~10⁵ samples
- Kernel choice and regularisation are dataset-specific
- Hard to extend to time-series without explicit windowing