Data quality measure
A specific way of measuring how good data is, for example, the percentage of records with a complete address field, or the proportion of duplicate entries.
The Simple Version
A specific way of measuring how good data is, for example, the percentage of records with a complete address field, or the proportion of duplicate entries.
Detailed Explanation
ISO/IEC 5259 Part 2 provides a framework for defining and applying data quality measures to AI data. Measures are associated with one or more quality dimensions (accuracy, completeness, consistency, etc.) and are specified in terms of the measurement function, the data element or dataset to which it applies, and the target or threshold value. In AI applications, measures may include completeness rates for feature vectors, label agreement rates across annotators, demographic representation ratios, or temporal coverage metrics. Measures are the operational tools that turn abstract quality dimensions into assessable, actionable indicators.
Key Characteristics
- Associated with one or more data quality dimensions
- Defined with a measurement function, scope, and acceptability threshold
- Can be applied to individual fields, records, or full datasets
- Forms the basis for data quality reporting and improvement prioritisation
Why It Matters
Data and AI teams should define a set of standard quality measures for each key training dataset, making quality assessments reproducible and comparable across dataset versions and AI model iterations.
Real-World Analogy
Like clinical trial endpoints, specific, pre-defined metrics (blood pressure reduction, survival rate at 12 months) against which a treatment's effectiveness is assessed, not vague statements of 'good health outcomes'.
Common Misconceptions
- A data quality measure is the same as a quality dimension, a dimension is a quality characteristic (e.g. completeness); a measure is the specific metric used to assess it (e.g. percentage of non-null values).
- One measure per dataset is sufficient, complex datasets may require dozens of measures across multiple dimensions to provide a complete quality profile.