Predicting Spatiotemporal Counts of Opioid-related Fatal Overdoses via Zero-Inflated Gaussian Processes

Abstract

Quals talk:

Overdose has recently become the leading cause of accidental death in the United States, ahead of vehicle crashes and gun violence. Working with public health collaborators, we provide a machine learning model with an intervention-aware performance metric that can address this problem by guiding the investment of limited intervention resources. This talk will explore a zero-inflated Gaussian Process (GP) model that can learn from historical death records to predict near-term risk of future opioid-related overdose deaths in all 1620 census tracts across the state of Massachusetts. We find zero- inflated GPs can prioritize regions in need of near-term public health interventions better than alternative models at finer spatial and temporal resolutions than most prior efforts. We will discuss an extension of this model to use a zero-inflated Poisson likelihood that is more conceptually appropriate to predict count data, preliminary findings, and future directions for this research.

Please join meeting in Cummings 280.

Zoom is not available for this event; please disregard dial-in passcode included in email.