An outlier detection algorithm based on a soft hyper-sphere for high dimension nonlinear data

In process industries, such as metallurgy and chemistry, real procedure parameters usually possess high-dimensional nonlinear features. To solve the problem of outlier detection in complex high-dimensional data, the concept of a soft hyper-sphere is introduced in this paper. An original data set is projected into a high-dimensional feature space using a nonlinear kernel function, and the boundary of the soft hyper-sphere is determined within this feature space. To avoid a mass product quality incident, location information on the testing samples, which are projected into the feature space, is used to decide whether they are outliers. As an applied example, practical procedure data obtained from a type of auto steel product were tested. The results verify that the proposed outlier detection algorithm based on a soft hyper-sphere has a better ability for outlier detection in high-dimensional nonlinear data than tradional methods.