Skip to content
    Flow-alerta.cab

    Equalized Odds

    Application & enablementEqualised oddsConditional error rate parity
    Editorially reviewed on: 1 source
    Your reading view

    The short answer · 14 words

    In short

    In binary classification, equalized odds requires equal true-positive and false-positive rates between considered groups.

    Flow-alerta.cab (2026): Equalized Odds. https://flow-alerta.cab/en/glossar/equalized-odds. Editorially reviewed on: 10 October 2026.

    How much detail would you like?

    Practical

    The metric makes group-specific error differences visible, but does not guarantee overall fairness.

    Hypothetical example: A selection model is tested for TPR and FPR per group, in addition to data quality and legal consequences.

    Companies can use it as a diagnosis in high-risk classification processes, not as automatic certification.

    At a glance

    Core
    Within each actual target class, the prediction should be independent of protected group membership. The criterion requires ground truth and is only one fairness measure.
    Use
    The metric makes group-specific error differences visible, but does not guarantee overall fairness.
    Example
    Hypothetical example: A selection model is tested for TPR and FPR per group, in addition to data quality and legal consequences.
    Limits
    Treat inaccurate labels as truth; ignore small groups; optimize metrics in isolation; use sensitive data without legal review.

    In context

    Demographic Parity

    Demographic parity compares whether positive model decisions occur equally or sufficiently similarly in protected groups.

    AI Fairness

    AI fairness explores how data, models and processes distribute benefits, errors and burdens between people.

    Questions about this term

    How is it different from demographic parity?

    Equalized odds conditioned to the actual target label.

    Is it achievable?

    Even trivial classifiers can fulfill it. A useful solution can entail conflicts of objectives and quality losses; the measure does not guarantee overall fairness.

    All details and background

    Within each actual target class, the prediction should be independent of protected group membership. The criterion requires ground truth and is only one fairness measure.

    ELI5 – Explained simply

    In actual positive and negative cases, groups should be treated equally often right or wrong.

    Why it matters in practice

    The metric makes group-specific error differences visible, but does not guarantee overall fairness.

    Example

    Hypothetical example: A selection model is tested for TPR and FPR per group, in addition to data quality and legal consequences.

    Pitfalls and limits

    • Treat inaccurate labels as truth; ignore small groups; optimize metrics in isolation; use sensitive data without legal review.

    Origin & history

    Hardt, Price and Srebro formulated the 2016 criterion for supervised learning systems.

    Use in mid-sized companies

    • Companies can use it as a diagnosis in high-risk classification processes, not as automatic certification.

    Related terms

    Matching Flow-alerta service

    AI opportunity analysis & roadmapAI consulting: Examine the suitability, data and prerequisites of a possible use case on the subject of “Equalized Odds”.

    Matching tool / use case

    View use casesUse Case: Group-related error analysis with professional and legal evaluation

    Sources

    Quellenprüfung: 10.10.2026. Definition und dokumentierte Einordnung abgeglichen; Beispiele beschreiben mögliche Anwendungen.

    Editorially reviewed on:

    Next step

    Put knowledge into practice

    Discuss your projectView use cases