Applied Math Colloquium – Mansoor Haider (NCSU)

Unsupervised Learning Approaches for Dual-Domain Clustering with Applications in Public Health

Abstract: Many policymaking questions in public health are driven by data sets representing both application-specific attributes and geographic location. Unsupervised machine learning techniques, such as clustering, enable features in the data to be learned without prior knowledge of the underlying mechanisms giving rise to the data. In dual-domain clustering, attribute data is augmented with geographic location, when assembling the data vectors prior to clustering. This approach aims to yield geographically cohesive clusters that also exhibit significant differences in attributes between pairs of clusters. The underlying clustering technique can be tailored, in part, by using a non-Euclidean distance with hyperparameters that provide a relative weighting of the geographic and attribute subspaces. This talk will present tailored dual-domain clustering methods for county-level data in two public health applications. The first application considers colorectal cancer incidence data in the state of California, with an emphasis on identifying disease disparities across demographic groups. The second application considers an extension of the dual-domain approach to (top-down) clustering of spatiotemporal data for Covid-19 infectious disease data in North Carolina, focusing on county-level disease hotspot detection. Here, a data-driven (bottom-up) ground-truth mechanistic SEIR model is also developed and calibrated using both county-level and statewide daily infection data during an outbreak. The tailored dual-domain clustering approaches presented in this talk are well-suited to public health or public policy applications that require effective coordination between municipal, county and state officials.

Speaker Bio: Mansoor Haider is a Professor in the Department of Mathematics and the Biomathematics Graduate Program at North Carolina State University. He is also Director of the Foundations of Data Science MS Program and an Associate Director of the Data-Enabled Modeling research unit in the Comparative Medicine Institute, both at NC State. His research in applied and computational mathematics focuses on mathematical modeling of biological soft tissues including biomechanics, mechanobiology, tissue engineering and wound healing, and on unsupervised machine learning with applications in the life sciences. Mansoor received his PhD in Mathematical Sciences from Rensselaer Polytechnic Institute in 1996, followed by three years as an NSF Mathematical Sciences Postdoctoral Research Fellow at Duke University. He joined the faculty at NC State as an Assistant Professor in 1999, served as Director of graduate programs in mathematics and applied mathematics from 2012-2016, and as an Associate Director at SAMSI from 2018-2021. He is a prior recipient of the ASME Richard Skalak (best paper) award as well as awards at NC State for outstanding teaching, and excellence in teaching and learning with technology.

Applied Mathematics Colloquium