{"id":9561,"date":"2026-07-27T12:30:27","date_gmt":"2026-07-27T10:30:27","guid":{"rendered":"https:\/\/www.kubicek.ai\/?post_type=lexicon&#038;p=9561"},"modified":"2026-07-27T13:29:31","modified_gmt":"2026-07-27T11:29:31","slug":"anomaly-detection","status":"publish","type":"lexicon","link":"https:\/\/www.kubicek.ai\/en\/lexicon\/anomaly-detection\/","title":{"rendered":"Anomaly detection"},"content":{"rendered":"<p class=\"wp-block-paragraph\"><strong>Anomaly detection<\/strong> is the task of finding samples that differ markedly from the majority behaviour of the data. It differs fundamentally from ordinary classification in that anomalies are rare, diverse and unknown in advance \u2013 so no representative training set of &#8220;all possible faults&#8221; can be assembled. The usual approach is therefore to learn from normal data only and to define an anomaly as whatever does not fit the learned model of normality. There are several groups of methods. Statistical ones start from deviation from an estimated distribution. Density- and distance-based methods, for example the local outlier factor, look for points in sparse regions. Isolation Forest exploits the observation that an outlying point can be separated from the rest of the data by random cuts in few steps. Reconstruction approaches using an autoencoder flag as anomalous any sample with a high reconstruction error. Practical pitfalls are defining normality in strongly seasonal data, the curse of dimensionality, and above all the sensitivity setting: an over-eager detector buries operators in false alarms and the system stops being used. Typical areas are fraud, predictive maintenance, cybersecurity and quality control.<\/p>\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n<p class=\"wp-block-paragraph\">An experienced ward nurse knows something is wrong with a patient before the machines sound an alarm. She does not know the list of every disease in the world \u2013 but she knows very well what a patient recovering normally looks like. And she senses a departure from that normality. Anomaly detection works exactly like this: it does not try to learn every possible fault, because there are infinitely many and you have never seen most of them. It learns only the usual and flags anything that does not fit. Its greatest danger is the same as with an oversensitive alarm \u2013 when it beeps constantly, people stop hearing it.<\/p>\n","protected":false},"featured_media":0,"template":"","class_list":["post-9561","lexicon","type-lexicon","status-publish","hentry"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.kubicek.ai\/en\/wp-json\/wp\/v2\/lexicon\/9561","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.kubicek.ai\/en\/wp-json\/wp\/v2\/lexicon"}],"about":[{"href":"https:\/\/www.kubicek.ai\/en\/wp-json\/wp\/v2\/types\/lexicon"}],"wp:attachment":[{"href":"https:\/\/www.kubicek.ai\/en\/wp-json\/wp\/v2\/media?parent=9561"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}