Simulate Controlled Outliers.
SCOUTerRpack
SCOUTer package in R
Simulating anomalous data is an extremely common procedure. However, very little attention is paid to this step and it is usually defined ad hoc, existing a lack of standard. In this package, a new framework to simulate outliers directly controlling their outlying properties has been proposed. This framework offers the possibility of generating data sets with all type of desired properties, given that the user can control the pair of statistics that essentially define outliers: the Squared Prediction Error ( SPE ) and the Hotelling-T2 ( T2 ). These metrics evaluate in a complementary way how far is an observation from the majority of a data set. Since Given an observation with initial values for the statistics, a PCA model and target values for the statistics, our simulation method drifts the observation in a direction that shifts the initial SPE and the T2 until reaching their target values.