How to simulate outliers with the desired properties

Deviating multivariate observations are used typically to test the performance of outlier detection methods. Yet, generation outlying cases itself usually appears as a secondary methodological step in methods comparison. In literature, outliers defined using certain distribution parameters which differ from those clean or reference data. However, these change among authors, leading lack standard and measurable definition characteristics simulated outliers. This makes comparison between hard its results dependent on procedure followed simulate order set procedure, framework is here. Since it based specifications for both Squared Prediction Error (SPE) Hotelling’s T2 statistics Principal Component Analysis (PCA) model, tuning them becomes simple efficient task. has been implemented Matlab functions.