Anomaly detection is the process of identifying observations that do not follow the normal patterns represented in data. Anomalies are also called outliers or abnormal observations, although these terms may have different meanings depending on the application. Anomalies can be point anomalies, contextual anomalies, or collective anomalies. A point anomaly is unusual on its own, a contextual anomaly is unusual within a particular context, and a collective anomaly involves an unusual group or sequence of observations. Detection methods include statistical thresholds, distance-based approaches, density-based methods such as Local Outlier Factor, isolation-based methods such as Isolation Forest, and reconstruction-based approaches such as autoencoders. Some systems learn from normal data, while others use labeled examples of both normal and anomalous behavior. Because anomalies are often rare, evaluation may require precision, recall, careful threshold selection, and expert review.
Anomaly Detection
Anomaly detection identifies observations, events, or patterns that differ significantly from the expected behavior of a dataset or system.
Explanation
Example or application
A banking system monitors transaction amounts and behavior. A transaction that differs substantially from a customer's usual activity may be flagged for further fraud investigation.
Text
Anomalies are unusual observations that may indicate errors, rare events, unexpected behavior, or potential threats.
Key Points
Anomaly detection identifies unusual observations or patterns.
Anomalies can be point, contextual, or collective.
Detection methods include statistical, distance-based, density-based, and isolation-based approaches.
Anomaly detection can be supervised, semi-supervised, or unsupervised.
Rare anomalies can make evaluation difficult.
A flagged anomaly is not automatically an error or fraudulent event.
Applications
Financial fraud detection
Network intrusion monitoring
Industrial equipment monitoring
Healthcare monitoring
Quality control
Learning Outcomes
Define anomaly detection.
Distinguish point, contextual, and collective anomalies.
Identify common anomaly detection techniques.
Explain why rare events create evaluation challenges.
Practice Questions
What is an anomaly?
How does a contextual anomaly differ from a point anomaly?
What is Isolation Forest?
Why can accuracy be misleading for anomaly detection?
Related Concepts
Outlier Detection
Isolation Forest
Local Outlier Factor
Fraud Detection
Imbalanced Data
Common Mistakes
Assuming every outlier is an error.
Using accuracy as the only evaluation metric for highly imbalanced data.
Ignoring contextual information.
Choosing detection thresholds without considering the cost of false alarms and missed anomalies.
Sources and verification
Sources are shown as provided in the project data. Not every source has yet been independently verified. Please report errors or missing citations.
- Chandola, V., Banerjee, A., and Kumar, V. (2009). Anomaly Detection: A Survey.
- Liu, F. T., Ting, K. M., and Zhou, Z.-H. (2008). Isolation Forest.
Suggest a correction or source · More in Artificial Intelligence & Machine Learning