A formal and integrated approach to engineering machine learning processes: A method base for project management

Abstract

Listen

Translate article

Enhancing project management (PM) for machine learning (ML) requires structured acquisition and application of PM knowledge. However, significant differences exist between managing ML-enabled software products (MLESP) and traditional software products (TSP). In modern tool-centric ML environments, creating a method base to support team learning and knowledge management is challenging. Studies also show that a “one-size-fits-all” approach to PM can fail to meet diverse team and organizational requirements. Indeed, the main challenge is capturing, storing, and reusing tacit knowledge on PM methods, processes, tasks, and tools for ML. The experimental, data-driven nature of ML may often lead to ad hoc processes, complicating integration with traditional software lifecycles. Therefore, tailoring a PM method for MLESP becomes critical. This study uses a mixed research approach combining Design Science Research (DSR), PM, Method Engineering (ME), and Process Algebra (PA). Key outputs include an ME framework for PM, a method base for ML, and a hybrid ML PM method tailored for Baskent University Hospital Ankara (BUHA). A use case-based scenario analysis technique validated the requirements phase of the hybrid ML PM method in the context of BUHA. The proposed approach can offer comprehensive, yet pragmatic and adaptable solutions as it blends the strengths of ML, PM, ME, and PA knowledge domains. Moreover, PA contributes formal and mathematical foundations for specifying and validating PM methods and tailoring processes. This study has the potential to contribute not only to ML PM and BUHA but also to advancing process management within the mission and safety-critical domains like healthcare.