Analyzing environmental and exercise influences on geriatric health via video data
Abstract
The growing global aging population necessitates a deeper understanding of the interactions between environmental and physical activity factors on geriatric health outcomes. Traditional methodologies predominantly utilize static clinical data, which often fail to capture the dynamic and complex relationships between environmental exposures and activity patterns in older adults. To address these constraints, we introduce an innovative approach that utilizes visual recordings to examine the relationship between surrounding conditions and physical activity patterns concerning elderly well-being. This framework comprises three core components: a symbolic representation of geriatric health variables, the Adaptive Geriatric Embedding Network (AGE-Net), and the Geronto-Actuator intervention strategy. The symbolic representation systematizes biomedical, behavioral, and environmental variables into a structured manifold, enabling the modeling of spatiotemporal dynamics inherent in aging populations. AGE-Net integrates multimodal data, including longitudinal medical records and frailty scores, through a low-rank structured representation that incorporates clinically significant constraints, facilitating interpretable health trajectory modeling. The Geronto-Actuator employs reinforcement learning to optimize patient-specific health action recommendations, targeting outcomes such as reduced hospitalization risk and stabilized functional status. Experimental results demonstrate the superiority of this framework in predicting adverse health events, offering a scalable and clinically relevant approach to advancing eldercare analytics and personalized interventions.