Into nature and you can type of anomalies: a look at deviations when you look at the analysis
Anomalies was situations during the an excellent dataset which might be somehow uncommon plus don’t match the general habits. The concept of the latest anomaly is typically ill defined and understood since vague and you can domain name-founded. More over, even after particular 250 numerous years of e-books on the subject, zero total and tangible overviews of your own different varieties of defects possess hitherto been wrote. By means of a thorough literary works comment this study therefore offers the first technically principled and you will domain-independent typology of information defects and gift ideas a complete report about anomaly designs and you can subtypes. So you can concretely explain the concept of brand new anomaly as well as different signs, the brand new typology utilizes four proportions: investigation variety of, cardinality off matchmaking, anomaly height, study structure, and analysis shipments. These types of simple and you will data-centric dimensions however give 3 wide organizations, 9 first items, and 63 subtypes out-of defects. New typology encourages the latest testing of your practical prospective away from anomaly detection algorithms, leads to explainable analysis research, and offers information towards the related subjects particularly regional versus worldwide defects.
