ISO-IEC 5259-4:2024

A common framework for data quality in ML

The ISO/IEC 5259-4 standard provides a framework for ensuring data quality in training and evaluation processes for analytics and machine learning (ML). It establishes general organizational approaches that are universally applicable, regardless of an organization's type, size, or sector.

This standard includes comprehensive guidance on the data quality process for various ML methodologies, such as:

• Supervised ML: Emphasizing best practices for labelling data used to train ML systems, including common organizational approaches to training data labelling.

• Unsupervised ML: Ensuring data integrity for applications without labelled data.

•  Semi-supervised ML: Combining labelled and unlabelled data effectively.

•  Reinforcement Learning: Managing data derived from interaction-driven methods.

•  Analytics: Enhancing data quality for analytical applications.

ISO/IEC 5259-4 outlines processes that can be applied to training and evaluation data from various sources. These processes cover data acquisition, composition, preparation, labelling, evaluation, and use, ensuring a comprehensive approach to data quality management. Importantly, the standard does not prescribe specific services, platforms, or tools, making it adaptable to diverse organizational needs.

Benefits of Using ISO/IEC 5259-4

Adopting ISO/IEC 5259-4 brings numerous benefits to organizations striving for excellence in analytics and machine learning. By establishing standardized approaches to data quality, this standard helps organizations reduce errors and biases in their training and evaluation data, ensuring that machine learning models perform reliably and accurately. Improved data quality not only enhances the effectiveness of analytics and ML systems but also builds trust with stakeholders by demonstrating a commitment to responsible and transparent data practices.

Organizations that implement this standard can streamline their workflows, reducing inefficiencies and saving valuable time and resources during data preparation and evaluation. Moreover, the guidance provided is adaptable to diverse organizational

contexts, making it easier for teams across industries to collaborate and maintain consistent quality standards. Compliance with ISO/IEC 5259-4 also positions organizations as leaders in data governance and innovation, helping them meet regulatory requirements, avoid costly rework, and ensure long-term scalability and success.