One-Class SVM Anomaly Boundary

This simulation visualizes how a one-class SVM separates normal data from anomalies without ever seeing a labelled anomaly during training. Every point contributes an RBF kernel "bump" of density; summed together those bumps form a decision surface, rendered here as a voxel volume, that encloses the bulk of the data. Points whose local density falls below a nu-controlled threshold sit outside the boundary and get flagged. Tune gamma to make the boundary tighter or looser, tune nu to control how aggressively it flags outliers, and switch between data shapes to see the boundary adapt to blobs, two separate clusters, or a ring manifold.