CS Colloquium: Forecasting County-Level COVID-19 Cases using Spatiotemporal Machine Learning, Social Media Connectedness, and Cell-Phone Movement Data

Abstract

Since the outset of the epidemic in the U.S., county-level forecasts of COVID-19 spread have been used for resource allocation and planning intervention strategies. Observed patterns of new cases/deaths as well as measurements of human interaction through proxies such as social connectedness or movement patterns have proved useful for predictive modeling of COVID-19. In this talk, I draw parallels between spatial statistical models and machine learning for capturing spatiotemporal dependence in the context of epidemiological data, and compare the power of Facebook’s social connectedness with cell phone-derived human mobility metrics for predicting county-level new cases of COVID-19. I outline our implementation of a SpatioTemporal autoregressive eXtreme Gradient Boosting (STXGB) model for forecasting county-level new cases of COVID-19 in the coterminous US. Evaluations show modest improvement in three- to four-week prediction horizons, when compared with a baseline Ensemble of 32-models currently used by the CDC. I will conclude with a discussion on the practicality, requirements, advantages, and disadvantages of using machine learning for forecasting the spread of infectious diseases, as well as opportunities for improving disease forecast and modeling.

Bio

Dr. Morteza Karimzadeh is an assistant professor of Geography and affiliate assistant professor of Computer Science at the University of Colorado (CU) Boulder. He joined CU from Purdue University, where he was a postdoctoral scientist at the School of Electrical and Computer Engineering. Morteza is a geospatial data scientist, with research cutting across geographic information retrieval, visual analytics, and spatiotemporal machine learning. His primary research focuses on method development, spanning various domains including social media analytics, scientific, environmental or public health data fusion and analysis, situational awareness, precision agriculture, and disease forecasting. His approach to research and development is human-centered, from visual design to ground truth creation, algorithm integration and evaluation, domain deployment and field studies. His other research project on sea ice classification and mapping is funded by a National Science Foundation EarthCube award, and he was recently awarded a RIO Innovation Seed grant to study the geographic patterns of neighborhood recovery post-pandemic.

https://cuboulder.zoom.us/j/190280621