Table 1 Model feature importance.
From: A prospective evaluation of AI-augmented epidemiology to forecast COVID-19 in the USA and Japan
USA model | Japan model | ||
|---|---|---|---|
Time-series features | Median rank | Time-series features | Median rank |
NPI schools | 1 | Mobility changes: residences | 1 |
NPI bar/restaurants | 2 | Std error of % of survey responders reporting CLI (unweighted) | 2 |
Snowfall (mm) | 3 | Estimated Reff | 3 |
Mobility index | 4 | Confirmed cases | 4 |
Cases/total tests | 5 | Cases mean-to-sum ratio | 5 |
NPI non-essential business | 6 | State of emergency | 6 |
Mobility samples | 7 | Mobility changes: transit | 7 |
Average temperature (°C) | 8 | Std error of % of survey responders reporting CLI (weighted) | 8 |
Confirmed deaths | 9 | % of survey responders reporting CLI (weighted) | 9 |
Cases mean-to-sum ratio | 10 | Mobility changes: workplaces | 10 |
Static features | Median rank | Static features | Median rank |
Ratio of the population over 60 | 1 | Average BMI of males | 1 |
Per capita income | 2 | Number of H1N1 cases in 2010 | 2 |
Mean air quality index | 3 | Number of new ICU beds | 3 |
Population density | 4 | Number of clinic beds/100 k population | 4 |
Number of households | 5 | Number of doctors/100 k population | 5 |
