Characterizing exposure to multiple air pollutants with a self-organizing map
Research by Brenna C. Kelly1,2,3, Simon C. Brewer2, and Michelle P. Debbink1,3
1University of Utah Department of Population Health Sciences; 2University of Utah School of Environment, Society, and Sustainability; 3University of Utah Department of Obstetrics and Gynecology
To better understand associations between multiple air pollutants and pregnancy complications, we trained a high-resolution neural network with weekly 1 km2 air quality estimates from a four-year period. This animation shows the training of a self-organizing map, a type of neural network which projects its nodes in two-dimensional space such that similar observations are close and dissimilar observations are distant. We linked patients to this surface based on their residence and timing of pregnancy, and this revealed that early-pregnancy exposure to mixtures of NO2, O3 and heat may be associated with preterm birth.
Attribution: This content was provided by researchers involved with the project.