COVID-19 Infection Trajectory
We built a machine learning model grounded in SIR dynamics, with key modifications to allow for granularity (at the county level) and much improved stability and predictive power.
Key benefits of our model
Mobility-Based Scenarios: Explicitly modeled dependence on dynamic social distancing metrics allows for forecasts based on various mobility scenarios.
County-Level Granularity: Key model parameters depend on county-level static and dynamic data including demographics, housing conditions, comorbidities, mobility data (social distancing adherence), and family interaction patterns. This yields granularity without over-fitting.
Our model includes counties above a certain threshold of total cases. Predictions are reported only for counties where the current difference between cumulative actual and predicted cases is less than 10% or 30 cases. The set of included counties represents over 90% of the total modeled cumulative cases.