A prediction is only useful in care if it can be trusted and understood. This work develops explainable models that forecast MCAS flares from continuous data while surfacing the reasoning behind each alert.
Black-box models can be accurate yet unusable in clinical settings where decisions must be justified. We prioritized interpretability alongside predictive performance for MCAS flare forecasting.
Using multi-parametric biosignal streams, we trained models constrained to produce human-readable feature attributions for every prediction.
Interpretable models matched black-box baselines within a small margin while providing per-alert explanations that clinicians rated as clinically plausible in 86% of cases.
Explanations let patients see which signals drove an alert, supporting shared decision-making rather than blind reliance on an algorithm.
Explainability and accuracy need not be at odds. Transparent flare prediction is a practical path to trustworthy AI in MCAS care.