Time-series forecasting

Time-series forecasting uses past and present observations to predict future biological quantities such as incidence, demand, population abundance or physiological measurements.

Temporal features

Models may use previous observations, moving summaries, seasonality, environmental variables and interventions as predictors.

Evaluation through time

Randomly mixing past and future observations between training and test sets can give unrealistic results. Forecast evaluation should preserve time order, training on earlier observations and testing on later ones.

Uncertainty

A useful forecast should communicate uncertainty as well as a central prediction. Forecast accuracy commonly decreases as the prediction horizon increases.