In progress

Development and Spatial Validation of a Random Forest Prediction Model for Firearm-Related Injury Risk in Chicago Census Tracts

Summary:
This project develops and spatially validates a random forest model predicting non-fatal firearm injury risk at the Chicago census tract level. Integrating spatial dependence via Moran’s Eigenvector Maps (MEMs) and a Bayesian spatial model, the analysis incorporates a comprehensive set of social, environmental, and structural predictors. The goal is to assess whether explicitly modeling spatial processes improves predictive accuracy and to identify high-risk neighborhoods shaped by unmeasured geographic processes such as disinvestment, segregation, and environmental hazards.

Research Questions:

• RQ1: Does incorporating spatial structure via MEMs and Bayesian spatial random effects improve predictive accuracy compared to non-spatial models?
• RQ2: Which neighborhoods exhibit elevated firearm injury risk due to unmeasured spatial processes, as captured by spatial random effects?
• RQ3: Does prediction error vary across geographic space, revealing limitations in model transferability?

Check out this Access our tutorial on spatial random forests here for more details.