RaPKod-package: Random Projection Kernel Outlier Detector

Description Details Author(s) References See Also


The RaPKod package implements a kernel method made for outlier detection. Namely, given a data set of reference typical observation (non-outliers or inliers), it tests each new observation in an online way to determine whether it is an outlier or not. This method uses random low-dimensional projections in a kernel space to build a test statistic whose asymptotic null-distribution (ie when the tested observation is not an outlier) is known. The RaPKod method has two parameters: gamma - the hyperparameter of the (Gaussian) kernel used - and p - the dimensionality of the random projection in the kernel space.


The package consists of two functions: the main function "rapkod" and the auxilary function "od.opt.param" which computes optimal parameters values in RaPKod.


Jeremie Kellner

Maintainer: Jeremie Kellner <jeremie.kellner@ed.univ-lille1.fr>


Kellner J., "Gaussian Models and Kernel Methods", PhD thesis, Universite des Sciences et Technologies de Lille (defended on December 1st, 2016)

See Also

rapkod, od.opt.param

RaPKod documentation built on May 2, 2019, 5:58 a.m.