Which of the following is not an example of an anomaly:
a person or thing that does not fit in; a peculiarity or strangeness: "A quiet man with the tastes and temperament of a girl is an anomaly in his exuberant family." —from Roget's 21st Century Thesaurus, Copyright
a deviation from what is normal or expected: "Anomalies are often linked to suspicious events or unexpected opportunities." —from an article by GE on anomaly detection.
Anomaly detection involves the search for unusual points or patterns in a set of data. It's a type of data mining technique used to find outliers (also known as standard deviations, outliers, noise, and exceptions). This process can be automated using machine learning training techniques. Anomaly detection is used to identify rare and interesting events in time series data – including hacking, fraud, crime, health problems, faulty equipment, and structural defects or infrastructure failures.
Anomaly detection is based on a model that compares input data against the prediction data set. If the difference between the predicted and the input data is larger than a threshold value, the algorithm flags the data point as an anomaly. The threshold value varies depending on the type of model used; for example, the threshold values in autoencoder and PCA models are the mean squared error (MSE), while the threshold in K-means models is the normalized distance. Anomaly detection algorithms may also use contamination values, which determine how much of the training data must be anomalous to flag a time series data point as an anomaly.
