Native JMP support for once-class SVM classifier or the SVDD implementation in SAS?

Question

Is it possible to do one class SVM classifiers for novelty detection natively in JMP? Something comparable to SAS SVDD (support vector data description) implementation. I have multivariate continuous data.

If not then are multivariate control charting techniques and outlier analysis tools sufficient for my purpose?

mjmg · Answer

Hi the SVM tutorials provided do not even remotely discuss my inquiry that's why I'm clarifying if there is a way to make this work with the current capabilities of SVM in JMP.

The intended use is the special case of unary/one-class SVM classification or outlier/novelty detection for unsupervised multivariate continuous data <https://en.wikipedia.org/wiki/One-class_classification> but the linked material deals with multiclass problems (supervised) with a predictor as a required parameter.

Here is the SAS implementation for reference:

https://documentation.sas.com/doc/en/pgmsascdc/9.4_3.2/casml/viyaml_svdd_overview.htm

The expected output could be a similar to outlier screening or a multivariate control chart.

ian_jmp · Answer

For the first question, try starting here. For the second here. But regarding 'sufficiency' that's hard to say without more details.