liveBook · Manning

Anomaly score

An anomaly score is a numerical measure of how unusual an item is relative to expected patterns or a specified reference.

How 3 Manning books explain it

Outlier Detection in Python

A numerical estimate of how unusual an item is relative to a specified reference group.

How it is used
Ranks records by estimated unusualness relative to a reference group, leaving any binary threshold to the application.
How deep it goes
Works through statistical, neighbor, histogram, and cluster-relative scoring, then examines score direction, thresholds, evaluation, and combination.
Examples
Uses fund segments, baseball nearest-neighbor scores, speech-data histograms, and FPOF and KDE score distributions.
What you can do after
Apply it via detector scores to rank investigation candidates, assess ranking quality, and choose thresholds cautiously.

Read how Outlier Detection in Python explains anomaly score →

Human-in-the-Loop Machine Learning: Active learning and annotation for human-centered AI

A numerical score indicating how strongly an item differs from patterns represented in reference data or a model.

How it is used
Selects unlabeled items for annotation by ranking model activations as outliers in an active-learning workflow.
How deep it goes
Calculates per-neuron ranks, interpolates between validation values, and compares averaging scores with using the lowest maximum logit.
Examples
Uses a 10-item activation ranking, a -0.36 activation scoring 0.9, and interpolation that gives -0.35 a score of 85%.
What you can do after
Apply it via model-based outlier sampling to prioritize unlabeled items for annotation.

Read how Human-in-the-Loop Machine Learning: Active learning and annotation for human-centered AI explains anomaly score →

Azure Security

Anomaly scoring is a web application firewall evaluation method that assigns a cumulative score to detected request anomalies to determine whether traffic is blocked.

How it is used
Determines whether Azure WAF on Application Gateway blocks requests that match managed rules.
How deep it goes
Explains severity values, cumulative addition, and the prevention threshold, but does not detail individual rule logic.
Examples
Uses critical, error, warning, and notice scores plus a two-rule total of 3 and 2.
What you can do after
Operate/configure prevention behavior by interpreting rule severities and predicting when requests will be blocked.

Read how Azure Security explains anomaly score →

Also covered in

What to know first

Related concepts

Questions about anomaly score

What is anomaly score?

An anomaly score is a numerical measure of how unusual an item is relative to expected patterns or a specified reference.

Which Manning books explain anomaly score?

Outlier Detection in Python, Human-in-the-Loop Machine Learning: Active learning and annotation for human-centered AI and Azure Security.

What should I know before learning anomaly score?

Manning books that teach anomaly score usually introduce probability distribution and outlier detection first.

These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.