Forecasting Methods

Summary Overview

There are several different methods we can use for forecasting each with its own pros/cons.

New cases

Providing the number of new cases expected for a given time period is important with respect to monitoring pandemic growth and tracking the pandemic.

Method	Pros	Cons
New cases (raw)	High granularity: In theory this would tell public officials exactly what days diagnosis will occur	<>
New cases 7 day rolling average	Removes the noise associated with reporting delays in data. Easier to see if model is learning disease trajectories	Hard to forecast for non-stationary data and predict out of distribution events
New cases difference	Still high granularity Works better on non-stationary data
7 day rolling average difference	Removes the noise associated with reporting delays. Helps solve non-stationary data problems	Less intuitive and comparable to other models Errors will compound

Hospitalizations

Forecasting the number of hospitalization helps with planning hospital staffing resources and monitoring how dire the situation. This is probably of greater utility to hospital officials than raw case numbers.

Method	Pros	Cons
New hospitalizations (raw)	High granularity: In theory tell public officials exactly what days diagnosis will occur	<>
Total active hospitalizations	Removes the noise associated with reporting delays. Easier to see if model is learning disease trajectories	Hard to forecast for non-stationary data and predict out of distribution events
Hospitalizations difference	Still high granularity Works better on non-stationary data
Active ICU admission	Still high granularity Works better on non-stationary data

ICU Beds

Analysis