Data Extraction Spreadsheet

Machine-learning-enabled systems are becoming increasingly common in different industries. Due to the impact of uncertainty and the pronounced role of data, the quality of such systems cannot be exhaus- tively characterized by existing quality models for traditional software. Machine-learning-enabled systems require a unique range of quality at- tributes, which can be achieved by the implementation of architectural tactics. Such decisions affect the further functioning of the system, its compliance with business goals, and they have to be made with a con- sideration of possible quality trade-offs. A related work analysis reveals the need to systematize the impacts of different architectural tactics on various quality attributes in the context of machine-learning-enabled sys- tems, since to the best of our knowledge, no systematic assessments of this issue have been performed yet. In this paper, to address this goal, we present a comprehensive study on the quality of machine-learning- enabled systems, architectural tactics, and their possible quality impacts. Based on a systematic literature review, we identified and categorized 24 quality attributes and 18 relevant architectural tactics together along with potential quality trade-offs. Our results systematize existing re- search as well as practical experience in building architectures of ML- enabled systems. They can be used by software engineers and researchers at the system design stage to assess the possible consequences of decisions made.