A Goal-Oriented Approach for Modeling Decisions in ML Processes

The ability to judiciously make systematic decisions throughout the entire Machine Learning (ML) lifecycle, encompassing model training, development, and deployment, is crucial for ensuring the robustness and adaptability of models in the face of evolving requirements. This paper proposes a goal-oriented (GO) approach which aims to improve the ML process so as to achieve overall ML goals. Our approach aims to understand the alternative types of decisions or choices that are made, and where in the ML process, based on what criteria. The main contribution of this research paper is to propose a GO approach, to support the strategic selection of technical alternatives, throughout the phases of model training, development, and deployment, to design ML solutions.