Forecasting
for Disease
Biosurveillance.
My research focuses on national-scale biosurveillance of Highly Pathogenic Avian Influenza (HPAI). This work has included classification and forecasting for the USDA ARS and is now continuing under a graduate research assistantship at Mississippi State University, with an emphasis on forward-looking forecasting, leakage-safe evaluation, and interpretable risk outputs.
Forecasting Framework
Modeled HPAI outbreak risk across the conterminous United States using environmental, agricultural, climatological, and geospatial data streams, plus historical outbreak activity and county adjacency structure.
Built forward-only forecasting infrastructure enforcing a strict temporal cutoff at each forecast origin. Unknown forecast-month covariates were synthesized from historical data, with time-gated feature construction, persisted forecast artifacts, and automated monthly backtesting.
Evaluated 40 combinations of forecasting backend, future-covariate strategy, and graph structure, including spatiotemporal and ensemble modeling.
Designed evaluation workflows for highly imbalanced data (~0.8% positives across 74,616 county-month forecasts), prioritizing balanced accuracy and recall, achieving 73% balanced accuracy and 74% recall, correctly identifying 417 of 563 outbreak-positive county-months.
Translated raw model outputs into HENCAST, an interactive dashboard converting county-level risk scores into Low/Medium/High biosurveillance tiers for poultry producers and animal-health professionals.
This work continues as part of my graduate research assistantship at Mississippi State University, advised by Dr. Isaac Jumper.