A Machine Learning Approach for Simulating the Airborne Transmission of Infectious Diseases
Disciplines
Aerodynamics and Fluid Mechanics | Artificial Intelligence and Robotics | Disease Modeling | Fluid Dynamics | Other Applied Mathematics | Other Computer Sciences | Other Mathematics
Abstract (300 words maximum)
Developing appropriate social protocols to prevent the transmission of infectious diseases, such as COVID-19, continues to be a persistent challenge across the globe. These challenges primarily stem from airborne transmissions, which tends to make infectious diseases far more widespread. Consequently, the information provided by epidemiological models is critical in the development of appropriate containment procedures. Common epidemiological models include the SEIR model, which examines the relationship between individuals who are traditionally categorized as susceptible, exposed, infected, or recovered. Incorporating the Wells-Riley equation can allow for these models to provide information with regards to infections within an enclosed space. However, there are inherent limitations to this approach that makes the predictions not accurate and limited in the information it provides. While more complex models using computational fluid dynamics can overcome these limitations, they are not as easily accessible due to their complexity and computationally intensive numerical calculations. With the advancement of machine learning techniques, there arises an opportunity to develop solutions which are more accurate, faster, and flexible than traditional techniques. These techniques typically employ neural networks which are trained by minimizing loss function errors to better fit the data. Through the utilization of deep artificial neural networks, this research presents a method for providing precise estimates to simulations which model the transmission of airborne particles within an enclosed space. Numerical results are used to compare for speed, accuracy, and to validate the proposed algorithm. The outcome of this research is the formation of a computationally efficient and flexible neural network framework which can be used in airborne transmission models, including those which incorporate the Wells-Riley equation.
Academic department under which the project should be listed
CCSE - Computer Science
Primary Investigator (PI) Name
Turaj Ashuri
Additional Faculty
(Note: I'm not 100% sure if mechanical engineering is the appropriate department for Dr. Turaj Ashuri since his department is listed as "Office of the Dean" in the directory)
A Machine Learning Approach for Simulating the Airborne Transmission of Infectious Diseases
Developing appropriate social protocols to prevent the transmission of infectious diseases, such as COVID-19, continues to be a persistent challenge across the globe. These challenges primarily stem from airborne transmissions, which tends to make infectious diseases far more widespread. Consequently, the information provided by epidemiological models is critical in the development of appropriate containment procedures. Common epidemiological models include the SEIR model, which examines the relationship between individuals who are traditionally categorized as susceptible, exposed, infected, or recovered. Incorporating the Wells-Riley equation can allow for these models to provide information with regards to infections within an enclosed space. However, there are inherent limitations to this approach that makes the predictions not accurate and limited in the information it provides. While more complex models using computational fluid dynamics can overcome these limitations, they are not as easily accessible due to their complexity and computationally intensive numerical calculations. With the advancement of machine learning techniques, there arises an opportunity to develop solutions which are more accurate, faster, and flexible than traditional techniques. These techniques typically employ neural networks which are trained by minimizing loss function errors to better fit the data. Through the utilization of deep artificial neural networks, this research presents a method for providing precise estimates to simulations which model the transmission of airborne particles within an enclosed space. Numerical results are used to compare for speed, accuracy, and to validate the proposed algorithm. The outcome of this research is the formation of a computationally efficient and flexible neural network framework which can be used in airborne transmission models, including those which incorporate the Wells-Riley equation.