One-Class SVM

A specialized kernel-based algorithm that learns a decision boundary around a single class of data to identify outliers and novelties in high-dimensional spaces.

One-Class SVM (OCSVM) is the standard for novelty detection when you only have normal data for training. Developed by Bernhard Schölkopf, the algorithm uses a kernel function (usually the Radial Basis Function) to map inputs into a high-dimensional feature space and separate them from the origin with a maximum margin. This creates a precise envelope around the training set. Engineers use it for critical failure detection: monitoring vibration data in jet engines, flagging credit card fraud, or identifying network intrusions. By adjusting the nu parameter (which sets the upper bound on training errors), you can fine-tune how aggressively the model labels new points as anomalies.

https://scikit-learn.org/stable/modules/outlier_detection.html#one-class-svm