A model-driven methodology for embedding AI bias mitigation requirements into SYSML: Application to smart home early chronic kidney disease (CKD) systems.

Abstract

Bias in AI systems poses risks of unfair outcomes, especially in sensitive domains such as healthcare. Existing approaches often address fairness only at the algorithmic level, leaving broader system-level safeguards under-specified. This paper presents a model-driven methodology for embedding AI bias requirements into SysML, enabling ethical concerns to be captured alongside functional and technical requirements. The methodology models safeguards such as confidence scoring, fairness evaluators, and human-in-the-loop oversight as firstclass SysML requirements, blocks, and interactions, ensuring traceability and verifiability within model-driven engineering processes. We demonstrate the approach through a smart home case study for early chronic kidney disease (CKD) monitoring, where system diagrams explicitly incorporate bias mitigation elements. This work contributes a structured method for operationalising ethical requirements in AI-enabled systems and lays the foundation for extending model-driven engineering with dedicated ethics modelling patterns.