Healthcare facilities face a critical challenge: elderly residents and patients with dementia experience heightened vulnerability to circadian disruption, yet practical methods to assess and optimize circadian light exposure remain limited by equipment costs, technical expertise requirements, and occupant compliance barriers. This paper presents a machine learning framework that predicts circadian stimulus (CS) exposure from readily observable environmental and behavioral parameters distance to window, blind angle, body position, and viewing direction without requiring wearable sensors or specialized measurement equipment. Developed using 512 spectral measurements from a nursing home environment over 13 weeks, ensemble models achieved test R2 values of 0.83 (Horizontal CS) and 0.78 (Vertical CS), demonstrating sufficient accuracy for environmental decision-making. This work directly addresses the practical barriers limiting circadian-aware care in healthcare facilities, providing a deployable tool accessible to facility staff without specialized training or capital investment, thereby supporting environmental optimization for vulnerable populations.
Beiglary et al. (Tue,) studied this question.