Identify & predict abnormal patterns in unbounded data streams.
A single instance of data is considered an anomaly, if it's too far off from the rest.
E.g.: Detecting credit card fraud based on the "amount spent" &/or the "transaction frequency"
This type of anomaly is more common in time-series data, where the abnormality is specific to the context.
E.g.: Spending $100 on food each day is considered normal during a holiday season, but may be considered odd otherwise.
This is a type of anomaly where a set of data instances collectively help in detecting anomalies. E.g.: An act of copying data from a remote machine to a local host unexpectedly, is anomaly that may be flagged as a potential cyberattack or data leakage.
IT & DevOps
Intrusion detection (system security, malware, etc.), monitor production systems and network traffic surges & drops, etc.
Banking & Insurance
Real time fraud detection (cards, insurance, etc.), stock market analysis, early detection &/or prediction of insider trading, etc.
Predictive maintenance, proactive product & process enhancements, service fraud and fault prediction & detection, etc.
Condition monitoring, early detection of seizures, tumors, diabetes, etc. and real-time knowledge sharing.