liveBook · Manning

Combined score

As explained in Outlier Detection in Python

An aggregate value produced from multiple detector scores for the same record.

A combined score aggregates scaled outputs from multiple detectors into one overall score for each record. The central decision is how strongly an extreme score should dominate, especially when detectors disagree. section 14.6

Agreement versus dominance

One record has four feature scores of 4.1, 4.2, 3.3, and 3.5; another has 15.1, 0.8, 0.3, and 1.1. The first shows broad unusualness, while the second depends on whether one extreme score should outweigh three ordinary scores. With multiple detectors, agreement among high scores increases confidence, but an isolated high score may represent either a missed anomaly or an unreliable detector. section 14.6

How Outlier Detection in Python covers it

How deep it goes
Defines aggregation and weighting issues but does not specify a combining formula.
Examples
Contrasts four feature-wise z-scores with four multivariate detector scores for two records.
What it assumes
Assumes scaled scores and a selected set of detectors.
What you can do after
Evaluate how detector agreement and extreme-score dominance affect an overall anomaly ranking.
How it connects
Connects outlier detectors, z-scores, multivariate detection, false positives, and false negatives.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Outlier Detection in Python. The section links point to the parts of the book it draws on.