Online One‐Class Classification With Least Squares Support Vectors

ABSTRACT Least squares support vector data description (LS‐SVDD) is a kernel machine with a closed‐form solution that produces a boundary for detecting multivariate outliers. Two closed‐form updates for either adding or removing a single vector from an LS‐SVDD model are derived, and, consequently, online algorithms for anomaly detection are proposed for streaming data. Moreover, a straightforward application for this online algorithm is a sequential multiple change‐point detection. Both algorithms are evaluated via simulation data, and the multiple change point detection algorithm is applied to a publicly available monitoring dataset involving wearable technology.