Percepção sobre requisitos de qualidade de dados para sistemas baseados em aprendizado de máquina: um estudo preliminar

Abstract

Machine Learning (ML) is a field of Artificial Intelligence that develops algorithms capable of learning from data and improving their performance through experience. With the increasing adoption of ML-based systems, data quality has become a critical factor. This quality is directly related to non-functional requirements (NFRs), which are essential for the effectiveness and reliability of ML models. However, applying NFRs in ML systems presents challenges, as some requirements become more critical, others change in meaning, and new ones may emerge. This study investigated developers’ perceptions of NFRs related to data quality in ML, identifying the most relevant requirements, the practices adopted to evaluate them, and the challenges faced. The research was conducted through a survey with professionals in the field, resulting in a diverse sample of developers, data scientists, and machine learning engineers. The results indicate that the main data quality attributes considered by professionals include consistency, accuracy, relevance, timeliness, and reliability. While participants have practical experience with data quality, the use of formal requirements engineering techniques in ML remains limited. The analysis revealed that data quality is often discussed in the early stages of development, but its continuous assessment throughout the system’s lifecycle still faces difficulties. The main challenges identified include the lack of standardized processes for evaluating data quality, difficulty in defining objective metrics, and the absence of adequate tools for continuous validation. Furthermore, the study highlighted that many professionals recognize the importance of NFRs for model quality, but there are still gaps in the systematic adoption of these practices in ML system development.