Procedure Decks for Transparency in Prescriptive Machine Learning

Abstract

To convey essential information about machine learning (ML) systems and the datasets and models they rely on, specialised documentation approaches have been created. A descriptive approach has been used mostly in techniques like Datasheets, AI FactSheets, and Model Cards, which provide numerous facts about the system's constituent parts. While the aforementioned data is crucial for product developers and outside experts to determine if the ML system satisfies their needs, other stakeholders may not find it to be as useful. To correct faults or enhance system performance, ML developers in particular want instruction on how to reduce potential flaws. We provide a documentation item with the intention of offering such direction in a prescriptive manner.