Data quality dimension
One of the key aspects of data quality, like accuracy (is it correct?) or completeness (is it all there?), used to frame what makes data good or bad for a given purpose.
The Simple Version
One of the key aspects of data quality, like accuracy (is it correct?) or completeness (is it all there?), used to frame what makes data good or bad for a given purpose.
Detailed Explanation
ISO/IEC 5259 Part 1 defines and organises data quality dimensions relevant to AI applications. Core dimensions include accuracy (correctness relative to a reference), completeness (absence of missing values), consistency (absence of contradictions), currentness/timeliness (data is not stale), and representativeness (data reflects the population to which it will be applied). Different AI applications prioritise different dimensions, a fraud detection model may prioritise recency, while a demographic prediction model may prioritise representativeness. The EU AI Act references quality dimensions implicitly in its data governance requirements for high-risk systems.
Key Characteristics
- Provides a structured vocabulary for specifying and communicating quality requirements
- Different dimensions matter more for different AI use cases
- Standardised in ISO/IEC 5259-1 for AI data quality contexts
- Forms the conceptual layer above individual quality measures
Why It Matters
AI product teams should specify required data quality dimensions as acceptance criteria when commissioning training data, preventing quality problems from being discovered only after model development begins.
Real-World Analogy
Like nutritional dimensions on a food label (calories, protein, fat, vitamins), each dimension captures a different aspect of the food's quality profile, and which dimensions matter most depends on the dietary goal.
Common Misconceptions
- All quality dimensions are equally important for every dataset, which dimensions are critical depends on the intended use and the risks of the AI application.
- Quality dimensions are fixed. ISO standards define core dimensions, but organisations can extend them with domain-specific characteristics relevant to their AI applications.