Teste de software em MLOps: um mapeamento sistemático da literatura

Abstract

This work aimed to analyze software testing practices in the context of MLOps, seeking to understand how to ensure quality and reliability in systems that integrate machine learning. The justification for this research lies in the growing adoption of MLOps and the need for testing methods adapted to this scenario. The methodology employed was a systematic mapping of the literature, conducted with well-defined selection criteria, in order to identify relevant studies on the topic. The analysis enabled the categorization of the types of tests applied, as well as the systematization of the reported practices. The results revealed that, although there are consistent initiatives, important gaps still remain, mainly in the definition of standards and in automated support for testing. It is concluded that the field presents opportunities for advancement both in research and in practice, especially in the development of more robust tools and guidelines for software testing in MLOps pipelines.