Severe acute malnutrition (SAM) can lead to death when untreated. Its timely identification and treatment is essential to prevent that fatal outcome. To that end, it is paramount to have an early information on its future evolution, in order to inform evidence-based and strategic decision-making.
Current (and traditional) approaches for estimating programme caseload, do not provide useful and actionable insights on the variation of the estimated caseloads across months, by showing its trend and seasonal variation (if existent). The alluded approaches are limited to provide a one-year-and-static information on the expected caseload. This limits the ability of programme managers to take strategic decision for times during the year when cases are expected to rise or fall.
In this repo, I try to address this gap by forecasting the SAM caseload in Somalia between January and December 2025, based on the information available in the historical SAM admissions from the past 72 months (5 years). The model does not try to account for new external inputs (dynamic or adaptive forecasting) or to find any drivers (explanatory forecasting). This is a pure time series model.
Forecasts were done for a horizon of 12 months, from January to December 2025, and they are split into the four main Somalia livelihood systems. A glance at the forecasted SAM cases in one livelihood system is shown below:
The above results seem to provide actionable insights for anticipatory action, which could include resource mobilization and scaling up of outreach activities over the months when SAM is expected to increase. Forecasting acute malnutrition through time series offers several advantages:
- it allows to anticipate the rise in the number of cases.
- it supports resource mobilization and allocation.
- It strengthens early warning systems.
- And it helps to save lives.
raw-data/: a data frame of the input data. This is encrypted.R/: some handy user-defined functions.scripts/: a set ofRscripts. These are split into different files, based on the specific task that they execute.
The following workflow is recommended:
flowchart LR
A(Retrieve secret key for decryption)
B(Load project-specific functions.R)
C(Run read-in-data.R)
D(Run data-wrangling.R)
E(Run exploratory-data-analysis.R)
F(Run decomposition.R)
G(Run split-training-test-data.R)
H(Run training-test-forecast-*.R)
A --> B --> C --> D --> E --> F --> G --> H
The above flowchart can be implemented simply by running the scrip.R
file found in the root directory.
The repository was created in R version 4.5.1 This project uses the
{renv} framework to record R package dependencies and versions.
Packages and versions used are recorded in renv.lock and code used to
manage dependencies is in renv/ and other files in the root project
directory. On starting an R session in the working directory, run
renv::restore() to install R package dependencies.
This project uses {cyphr} to encrypt the raw data that lives in
data-raw/ directory. In order to be able to access and decrypt the
encrypted data, the user will need to have created their own personal
SSH key and make a request to be added to the project. An easy-to-grasp
guide on how to make a request will be found
here.
This repository is licensed under a GNU General Public License 3 (GPL-3).
If you wish to give feedback, file an issue or seek support, kindly do so here.
Forecasting Seasonal Acute Malnutrition: Setting the Framework.
Tomás Zaba