A few anomalies regarding the trajectories off moving organizations is actually showed for the [118, cf
Barnett and you will Lewis [dos, cf. 30, 131] build an improvement anywhere between extreme however, genuine people in the main people, we.e., arbitrary movement on tails of the focal shipments, and you can contamination, which happen to be observations from a separate delivery.
Wainer distinguishes between distant outliers, and therefore showcase extreme values and generally are clearly in error, and you may fringeliers, which can be uncommon but with their reputation about three simple deviations on the majority of the data can’t be said to be really unusual and you can unequivocally erroneous. Essentially the same distinction is made in the that have white crows and you can in-disguise anomalies, correspondingly. Relatedly, inside [5, 133] a distinction is done anywhere between a failure outlier (noise) and an effective outlier (a significant departure of typical conclusion). The second classification is sandwich-split when you look at the events, i.age., uncommon changes in the real-community condition, and aspect problems, particularly a flawed detector [134, 135]. A total group was presented for the , towards the groups away from defects proving the underlying reasons for having the deviant nature: a proceeding mistake (elizabeth.grams., a coding error), a remarkable experience (instance a hurricane), an amazing observation (unexplained deviation), and you can yet another value combination (which includes typical thinking because of its private qualities). Most other present consider similar causes for the a very totally free-structure trend [39, 97, 136]. Inside a significant difference is created ranging from nine kind of anomalies. Some other wide class is the fact from , which distinguishes anywhere between around three general categories. A place anomaly makes reference to you to definitely or numerous personal cases you to definitely are deviant according to the remaining portion of the study. Good contextual anomaly looks normal in the beginning, but is deviant when a clearly selected perspective are taken into the account [cf. 137]. A good example try a temperature value that is merely surprisingly lower relating to the nazwa uЕјytkownika chatiw summer season. In the end, a collaborative anomaly makes reference to a set of data items that belong together with her and you may, due to the fact a group, deviate in the rest of the analysis.
Numerous certain and you may concrete classifications are identified, specifically those dedicated to sequence and you may chart data. A lot of their anomaly versions might possibly be demonstrated in more detail for the Sect. step 3. In the long run series research multiple within-succession designs are acknowledged, for instance the additive outlier, short term transform, peak change and you may innovational outlier [138,139,140,141, 191]. The brand new taxonomy demonstrated within the centers on between-succession defects inside the panel analysis and you can can make an improvement between remote outliers, shift outliers, amplitude outliers, and you may profile outliers. Various other certain classification known regarding regression study, where it is common to recognize between outliers, high-influence issues and you can important activities [3, 143,144,145]. 146, 147], namely the fresh new positional outlier, that’s operating out of a decreased-density area of the trajectory place, and angular outlier, which includes an instruction not the same as regular trajectories. The latest subfield from graph mining even offers acknowledged numerous specific categories regarding defects, with anomalous vertices, edges, and you can subgraphs as being the basic variations [18, 20, 112, 113, 148, 149]. Inside Sect. step three these defects, for example people who make it a data-centric definition, was chatted about in detail and you may positioned within this study’s typology.
Dining table step one summarizes the fresh new anomaly groups accepted on the extant literary works
The classifications in Dining table step one are either too general and you can conceptual to add a very clear and you may real comprehension of anomaly systems, or ability well-outlined designs which can be simply associated to own a specific objective (eg time collection studies, chart mining or regression acting). The brand new fifth-column in addition to produces obvious that extant overviews barely render obvious principles in order to systematically partition new classificatory place to acquire meaningful types of anomalies. They therefore don’t comprise a classification or typology as laid out by the . Towards better of my personal knowledge so it study’s structure and its particular predecessors offer the first complete typology from anomalies that presents a great total review of real anomaly systems.