Studying Software Architecture Design Challenges, Best Practices and Main Decisions for Machine Learning Systems

Abstract

Machine learning (ML) refers to statistical modeling techniques, which have recently sparked interest in the ML applications and service industries. The continuous usage of machine learning necessitates addressing software architecture (SA) design challenges and requires guidelines to overcome design issues. ML software system design in small and especially in big software engineering projects is a collaborative decision-making process in which software architect designers, researchers, and developers make design decisions. After considering various design alternatives, the development team handles design issues, examines the best design practices, and picks the main design decisions. In this paper, we provided the common challenges, best design practices, and major decisions for designing the software architecture of ML systems. The systematic literature approach (SLA), with snowballing, is used to extract academic papers from four databases. The inclusion/exclusion techniques helped to extract relevant articles according to research questions. Total 12 interviews were conducted from 9 countries across five continents with academic researchers and industrial professionals having a machine learning experience. Finally, SLR out comes and interview results are mapped. The mapped and unmapped data are discussed in this study to get more insight into the software architecture design decision for machine learning systems.