Table 2.

Machine learning US-CDC COVID-19 forecasting models- their description, features employed, methodology, and assumptions they make regarding public health interventions.

Model	Features used for forecasting	Author	Method	Assumptions
QJHong-Encounter (19)	Uses (1) Reproductive Number (R) and Encounter Density (D) relation in the past as a training set, (2) future D as input, and (3) ML/regression, the model predicts future R, and ultimately future daily new cases.	QJHong	Machine learning	Assumes current interventions will not change during the forecasted period.
OneQuietNight-ML (20)	Uses high-level features of daily case reports and movement trends data to make predictions about future Covid-19 cases.	OneQuietNight
JHU_CSSE-DECOM (21)	County-level, empirical model driven by mobility, epidemiological, demographic, and behavioral data.	JHU CSSE
UpstateSU-GRU (22)	A feed-forward RNN is used. The Seq2Seq algorithm trains the model to convert sequences from input to those in the output. The model inputs daily smoothed incident cases, deaths count, google mobility index, daily reproduction number, county demographic and health risk indices to model the baseline risk score.	SUNY Upstate and SU COVID-19 Prediction Team	County-level forecast using RNN seq2seq model.