Anomaly Detection
Unsupervised methods that surface cases which deviate from an established norm without needing labelled examples.
Uses AIestablishedunsupervised-ml
Anomaly detection earns its place when labelled outcomes are scarce or when the concern is a pattern nobody has described yet.
It is most valuable as a source of discovery, generating candidates for review rather than final decisions.
Treat its output as a research queue, not an action queue, because a high anomaly score only means a case is unusual relative to its peers.
It is the wrong choice whenever a decision needs to be justified case by case, since an anomaly score alone rarely explains itself.
General strengths
- Works without labelled outcomes.
- Surfaces novel patterns that no rule or model anticipated.
- Useful for finding candidates worth labelling for later supervised work.
General limitations
- Unusual is not the same as wrong, so precision is typically low.
- Explaining why a case scored highly takes extra work.
- Sensitive to how the population is segmented and features are scaled.
Applied to business problems
Each of these is written for its own problem. The technique is shared, the solution is not.
- Peer group behaviour outliersUnsupervised scoring that compares an account's activity against its peer segment to surface emerging laundering patterns before any rule or label exists for them.Transaction Monitoring
- Course engagement screening for personal tutorsA weekly screen of each student's activity in the learning management system and other campus systems against their course cohort, which flags students whose engagement has fallen far below their peers to their personal tutor or a support team, using no demographic inputs.Students Leaving Before They Finish Their Degree
- Outlier claim discoveryUnsupervised scoring that surfaces claims deviating from the norm for their segment, used to find new fraud patterns worth labelling.Claims Fraud Detection
- Sensor drift screeningUnsupervised monitoring of sensor reading distributions that flags when a stream has shifted, without asserting that the equipment itself is failing.Unplanned Equipment Downtime
- Phantom inventory screeningFlags products whose sales have stopped for longer than their demand history makes plausible while the inventory record still shows stock, and sends a store worker to check the shelf.Empty Shelves and Excess Stock
- Consumption anomaly screeningUnsupervised screening of smart meter consumption for premises whose pattern has changed or differs from similar customers, used where inspection history is thin to find candidates for inspection, many of which turn out to be faulty meters rather than theft.Electricity Theft and Non-Technical Losses