Gauging Size Resolved Ambient Particulate Matter Concentration Solely Using Biometric Observations: A Machine Learning and Causal Approach

Abstract

Human physiology responds to external environmental stimuli across different time scales. Long-term exposure to heat, air pollution, and volatile organic compounds has documented adverse effects on human health. Our previous work established that autonomic responses of the human body, collected from biometric sensors on a single subject, can successfully empirically model the concentration of particulate matter (PM) in the surrounding microenvironment. In this study, we demonstrate that this relationship holds across multiple observations involving multiple participants. Each subject rode a stationary bike outdoors while adorned with various biometric sensors, and the im- mediate environment around the subject was also captured through an array of environmental sensors. With additional data, Machine Learning (ML) models were able to predict PM concentrations (PM 2.5 at = 0.97) with only a handful of easily accessible biometric fea- tures, such as skin temperature, heart rate, and respiration rate. The causal perspective explored highlights the importance of physiological responses as indicators of exposure to particulate matter, providing a basis for the use of biometric data in environmental health surveillance and real-time pollution assessment.