COMPEL Glossary / anomaly-detection
Anomaly Detection
Anomaly detection is a technique that identifies data points, events, or patterns that deviate significantly from expected behavior.
What this means in practice
It is widely used in cybersecurity (unusual network traffic), fraud detection (suspicious transactions), manufacturing quality control (defective products), and infrastructure monitoring (equipment failures). Anomaly detection can use unsupervised or semi-supervised approaches, making it valuable when labeled examples of rare events are scarce. For organizations, anomaly detection systems serve as early warning mechanisms that can catch problems before they escalate. In the COMPEL framework, anomaly detection is also applied to AI systems themselves -- monitoring for model drift, data quality degradation, and unexpected behavioral changes in production AI.
Why it matters
Anomaly detection serves as an early warning system that catches problems before they escalate into costly incidents, making it valuable across cybersecurity, fraud prevention, quality control, and infrastructure monitoring. For organizations operating AI in production, anomaly detection applied to AI systems themselves — monitoring for model drift, data quality degradation, and behavioral changes — is essential for maintaining reliable, trustworthy AI operations.
How COMPEL uses it
Within the Technology pillar, anomaly detection is both an AI capability to be deployed and a governance mechanism applied to AI systems themselves. During Calibrate, existing anomaly detection capabilities are assessed. The Model stage designs monitoring architectures that include anomaly detection for AI system health. During Produce, detection systems are implemented, and the Evaluate stage reviews whether anomaly detection is catching degradation before business impact occurs.
Related Terms
Other glossary terms mentioned in this entry's definition and context.