Outlier Detection, as the name suggests, is to identify data that is different from normal data and data that differs greatly from expected behavior
. Identify problems such as credit card fraud, industrial production abnormalities, network flow abnormalities (network intrusion), etc., for a small number of incidents.
Point anomaly
Context anomaly
Group anomaly: For example, a collection of fake accounts in a social network is regarded as a group anomaly subset, but individual nodes in the subset may be as normal as real accounts.
Supervised: Both positive and negative examples of the
training set have labels.
Unsupervised: No labels on the training set. Semi-supervised: There are only instances of a single category (normal instances) in the training set, and no abnormal instances participate in the training.
Anomaly detection scenarios:
fault detection
, IoT anomaly detection,
fraud detection,
industrial anomaly detection,
time series anomaly detection,
video anomaly detection,
log anomaly detection,
medical daily detection,
network intrusion detection
Anomaly detection method:
statistical test: the effectiveness of statistical methods is highly dependent on the validity of the statistical model assumptions made for the given data.