From Indoor Environmental Exposure to Vascular Health Risk Prediction: An Interpretable Machine Learning Framework for Identifying Acute Cognitive and Microvascular Stress Responses

Abstract

Environmental exposures are increasingly recognized as contributors to endothelial dysfunction, microvascular impairment, and early vascular symptoms. Indoor air quality (IAQ), specifically exposure to particulate matter, carbon dioxide, and thermal stressors, has been shown to influence autonomic balance, inflammation, and cerebrovascular responsiveness. This study reframes a machine learning model originally designed to predict concentration difficulties as a proxy for early cerebrovascular stress. Using multimodal IAQ and perception data collected in a university building, an interpretable Random Forest Classifier achieved an AUC of 0.6167 in predicting acute cognitive difficulty—considered here as a functional marker of transient neurovascular strain. Explainable AI revealed that subjective perception of “stuffy” air was the strongest predictor, surpassing CO₂, PM₂.₅, and temperature. These findings align with emerging evidence linking environmental discomfort, autonomic dysregulation, and cerebrovascular perfusion changes. This work highlights the importance of integrating subjective environmental perception into vascular risk screening models and underscores XAI as essential for clinical and environmental health applications.