Recent studies report that a considerable amount of Machine Learning (ML) projects fails, and they lead to an inadequate return on investment or unsatisfactory results. There is also an overestimation of the opportunities that may be presented by ML projects. The majority of ML studies focus on technical aspects rather than project management (PM) issues. Therefore, one important factor has been adopting or customizing a PM method for the specific requirements of ML applications. More than 80% of ML engineers state that the PM methods, which can be tailored for ML, would improve the project performances. Traditional PM, Crisp-DM, Team Data Science Process, and Data-Driven Scrum are amongst the PM methods preferred for ML applications. Software engineering (SE) methods that are similar to the ones in the domain of SE, such as Scrum and Kanban, are also used in the industry. However, none of them is completely suitable for the idiosyncratic requirements of ML. Therefore, our approach to ML PM is based on the principles and guidelines of Situational Method Engineering (SME). As an engineering discipline, SME can allow building and adapting methods for SE or information systems development. Initially, specification of the domain-specific requirements of ML context is set as a method engineering goal. Later, an assembly-based approach is adopted, and thus, the method chunks existing in Crisp-DM, Team Data Science Process, Data-Driven Scrum, Scrum, and Kanban form the method bases. Finally, these method bases are combined to build an appropriate PM method for the ML context
Anahtar Kelimeler: Machine Learning, Project Management, Method Engineering, Situational Method Engineering
|
