A decision boundary is a surface separating regions assigned different decisions. In one-class outlier detection, an OCSVM fits a compact boundary around normal training data, allowing later points inside or outside it to be classified as normal or novel. section 5.4
The boundary balances compactness against excluding too many training points. Points just outside may still be normal, while points far outside are stronger outlier candidates; the boundary is therefore informative rather than infallible. section 5.4
Plotting an OCSVM boundary
Two synthetic two-dimensional clusters are used to fit an OCSVM. A plotted contour shows the learned boundary and distance from it, making normal and outlying test points visually comparable.
