Data quality
Data quality concerns whether data can support the purpose for which it is being used. A dataset may be accurate for accounting but insufficient for reconstructing a workflow. Useful assessment therefore combines technical checks with knowledge of the business process, intended task, and consequences of errors.
How it works
Define acceptance criteria, then examine missing values, invalid ranges, inconsistent units, duplicate records, label reliability, and coverage across relevant groups or periods. Check whether status meanings or collection methods changed over time. Compare samples with authoritative sources when permitted.
Document findings and remediation rather than silently replacing uncertain values. A valid outlier can be important evidence, while a normal-looking value can still be wrong.
Why it matters for licensing
A recipient needs to know what preparation remains and which conclusions the dataset can support. Quality evidence can inform feasibility and valuation, but it is not a guarantee of commercial demand or model improvement. Rights and privacy require separate assessment.
Example
Fictional example: A service export mixes durations recorded in minutes and hours. The team identifies the source-system change, normalizes the values with documented rules, and flags records whose units cannot be established.
Limitations and misconceptions
No single score captures every quality dimension. Improving one property can reduce another, such as removing rare cases to make a dataset more uniform. Checks should be tied to the intended task and repeated when the data or use changes.
Questions to ask
- What does fitness for this use mean in measurable terms?
- Which gaps, inconsistencies, or label errors affect the intended task?
- How are corrections and unresolved issues documented?
Sources
- Gebru et al. — Datasheets for Datasets · Accessed
- Google Developers — Machine Learning Glossary · Accessed
Explore whether your business data could be a fit.
Start with a description of your systems—not a data upload.