The SVM can be used to find a boundary of a model-free distribution in a multivariate sample space using the criterion (B. Schoelkopf et al., 2001)

This boundary is again a linear hyper plane in x and corresponds approximately to a ν-quantile of the distribution. The coefficient ν, 0<ν<=1 can be thought of as the ratio of the “outliers” in the data sample that will be outside of the distribution (and thus will not influence the model). The user however does not label any specific points to be excluded from the model. The model defines the “correct data” that belong to the distribution and can also be used to decide whether new points belong to this distribution. Robustness of this method makes it suitable for detecting stability and change points in complex multivariate processes with any type of distribution.