Statistical outlier labelling – a comparative study
Paweł D. Domański · 2020
Outliers always exist in real industrial data. They originate from different, often unknown sources. They are considered with respect in statistical analysis, robust regression and in data mining. One may find a lot of interesting approaches and consideration. Their detection, labelling, identification, isolation, filtering and interpretation is subject of many research activities. On the other hand effect of the outliers on control systems analysis has not been sufficiently investigated. Their effect is often neglected or considered contemptuously. This work addresses the subject of outlier detection from the perspective of control system performance analysis. The work focuses on statistical data-driven approaches. Selected statistical outlier detection approaches are proposed and compared on real industrial control loop data originating from process industry. Obtained results form a starting point for potential application of automatic outlier detection methods.