Overview
ISO/IEC 5259-4:2024 - Artificial intelligence - Data quality for analytics and machine learning (ML) - Part 4: Data quality process framework defines a general organizational framework to ensure data quality for training and evaluation across analytics and multiple ML paradigms. The standard applies regardless of organization type or size and covers data lifecycle stages from acquisition and composition through preparation, labelling, evaluation, provisioning and decommissioning. It explicitly applies to supervised, unsupervised, semi‑supervised and reinforcement learning as well as analytics, and does not prescribe specific services, platforms or tools.
Key Topics and Requirements
- Data Quality Process Framework (DQPF): Principles and a structured process for planning, evaluating, improving and validating data quality for ML and analytics.
- Data requirements & planning: Defining data needs for training and evaluation, dataset composition and provenance.
- Data acquisition & preparation: Best practices for sourcing, cleaning, transforming, encoding and de‑identifying data used in ML workflows.
- Data labelling & annotation: Guidance on labelling methods, labelling specifications, task assignment, process control, quality checking and revision-especially for supervised ML.
- ML‑specific processes: Tailored guidance for supervised, unsupervised, semi‑supervised and reinforcement learning, including recording and dataset handling.
- Data provisioning & decommissioning: Procedures for releasing datasets to model pipelines and retiring datasets safely.
- Roles of participants: Defined roles such as data planner, originator, collector, engineer, holder and user-to support accountability and process control.
- Assessment & improvement: Data quality assessment metrics and iterative improvement mechanisms; process validation to ensure fitness for purpose.
- Scope limitations: The standard addresses organizational approaches and processes, not particular tools or technical implementations.
Applications and Who Uses It
ISO/IEC 5259-4 is practical for organizations that build, evaluate or govern AI/ML systems, including:
- Data scientists & ML engineers designing training and evaluation datasets.
- Data engineers & platform teams implementing data pipelines, encoding and de‑identification.
- Data quality, governance & compliance officers establishing organizational controls and audit trails.
- Annotation vendors and labelling teams applying standardized labelling workflows and quality checks.
- Analytics teams ensuring reliable inputs for statistical analysis and business intelligence.
Adopting this standard helps reduce bias, improve model reliability, and support regulatory compliance by formalizing data quality processes across ML lifecycle stages.
Related Standards
- Part of the ISO/IEC 5259 series on AI data quality. ISO/IEC 5259-4:2024 complements other organizational and technical AI standards by focusing on the data quality process framework rather than tools or platforms.