When training data mostly describes normal cases, a one-class SVM estimates a region containing most observations. New observations outside that region may be flagged as anomalies or novel cases. section 8.8.1
Mathematically, it separates mapped training observations from the origin in feature space while allowing selected violations. This redirects support-vector ideas from separating two labeled classes toward detecting departures from an established pattern. section 8.8.1
One-class SVMs remain useful when normal observations are available but failures, defects, or intrusions are rare. Kernel methods can become computationally expensive as training data grows, and performance depends on choosing a suitable kernel and its parameters. section 8.8.2

