How can One-Class SVM be used for novelty detection?
Updated May 5, 2026
Short answer
It learns a hypersphere or boundary around 'normal' data[cite: 1].
Deep explanation
It maps data to a high-dimensional space and finds a hyperplane that separates the majority of data from the origin[cite: 1].
Real-world example
Detecting a new, unseen manufacturing defect[cite: 1].
Common mistakes
- Using it on datasets with many existing anomalies during training[cite: 1].
Follow-up questions
- Best kernel for One-Class SVM?