Handling Non-Fuctional Requirements in IoT-based Machine Learning Systems

In recent years, IoT and machine learning solutions have garnered immense popularity with the rise of industry 4.0. It is extremely important to document system requirements to deliver solutions that meet customer demands and increase satisfaction. However, the inclusion of non-functional requirements (NFRs) in software systems is either outright neglected or they are introduced in the later stages of the software development lifecycle, which often leads to architectural debt and, in the worst-case scenario, project failure. In the machine learning context, the poor handling of non-functional requirements affects architectural decision-making, which can potentially cripple the resulting system. Furthermore, our knowledge of existing NFRs no longer applies in this context due to the unique quality attribute concerns presented by the ML systems. This work aims to review the relationship between requirements engineering and software architecture w.r.t machine learning and present recently proposed methodologies for documenting and handling NFRs to deliver quality software systems. Three recent methodologies were also highlighted and compared.