Comparing with Other Models

We compare performance of major COVID-19 models to our Caltech CS156 Model in forecasting the pandemic (press article). Every day, our component models that had performed best in prior weeks are aggregated to create the CS156 model for predicting daily mortality.

Here is a comparison of the average performance, with more details below. Data is reported by The New York Times. The errors are computed statewide, and a perfect model corresponds to an RMSE of zero.

CDC Ensemble:

Compared to the CDC ensemble of 45 major models put together, the CS156 model by itself has been more accurate 52% of the time.




[All models trained with cutoff the day before the first prediction] Back to top