#######################################################################

This repository have 24 ARIMA and ARIMAX models which forecast short-term 
influenza-related hospitalizations in 48 states of the contiguous U.S. 
It also includes a CDC FluSight baseline model adapted to our work.

####

Our main folder has a Rmakdown that generates images and analysis for our 24 models
manuscript_results.Rmd
It also has a Rmarkdown which generates images and analysis comparing the models and the FLuSight baseline.
flusight_comparison.Rmd

#########################################################################
There are 4 subfolders in our main folder 

ERA5 Data/ (which has the code for treating the temperature and humidity data which we later use in some of our models and analysis)
models_with_logback/ (which has the models that use log_back transformations)
models_without_logback/ (which has the models that do not use log_back transformations)
Flusight-baseline-adapted/ (which has the CDC FluSight baseline which we adapted to our work)

######

In each of our models folders we have a Rmarkdown which process the NHSH hospitalizations dataset to a format that we can use (flu_hospitalization_dataset.Rmd). 
We also have a few additional datasets which we used for modeling or analysis (shapefiles and regression_features) 
which we decriped on our methods section.

Moreover, on each of our models folders we have 5 major model families each on a different subfolder.

ES_ARIMA/ Ensembles and single best ARIMA models. 
ES_TEMPERATURE/ Ensembles and single best ARIMAX models using 1-week-lag mean state temperature as covariate.
ES_EPIWEEK/ Ensembles and single best ARIMAX models using 1-week-lag mean cases by epidemiological week for the as covariate. Value calculated for the previous 2 years.
ES_ADJACENT/ Ensembles and single best ARIMAX models with 1-week-lag using mean hospitalizations in adjacent states as covariate.
ES_AVERAGE/ Ensembles and single best ARIMAX models with 1-week-lag using average of hospitalizations across the U.S as covariate.

#########

Each model family subfolder have the results of 3 model types (AUTO, ES27 and ES64) 
AUTO is a non-ensemble model which uses the best automatic fitting model. 
The other 2 model types are ensembles which combine 27 or 64 ARIMA or ARIMAX models. 
These results are loaded and evaluated in the manuscript_results.Rmd file
and on the flusight_comparison.Rmd file.

###########################################################################

[updated] 15Jan2026