| --- |
| license: mit |
| language: |
| - en |
| tags: |
| - climate |
| - health |
| - immunisation |
| - south-sudan |
| - flood-prediction |
| - xgboost |
| - random-forest |
| - lstm |
| metrics: |
| - auc |
| - f1 |
| pipeline_tag: tabular-classification |
| --- |
| |
| # Salama Health: Climate Disruption Index Models |
|
|
| Trained models for predicting climate-driven health facility disruptions in South Sudan, 28 days in advance. |
|
|
| ## What These Models Do |
|
|
| These models power the **Climate Disruption Index (CDI)**, a composite score between 0 and 1 assigned to each Primary Health Care Centre across Unity, Jonglei, and Upper Nile states in South Sudan. |
|
|
| The CDI estimates the probability that a facility will be cut off from the communities it serves due to flooding, road damage, cold chain failure, or population displacement within the next 28 days. |
|
|
| CDI scores feed into the **Immunisation Gap Score (IGS)** which ranks individual children by visit urgency for community health workers. |
|
|
| ## Performance |
|
|
| | Model | OOF AUC | Ensemble Weight | |
| |---|---|---| |
| | XGBoost | 1.0 | 0.341 | |
| | Random Forest | 0.9548 | 0.3256 | |
| | LSTM | 0.9775 | 0.3334 | |
| | Ensemble | nan | | |
|
|
| ## CDI Formula |
|
|
| CDI = 0.35 x P(flood) + 0.30 x P(cutoff) + 0.20 x P(CCF) + 0.15 x P(disp) |
|
|
| ## Pilot Coverage |
|
|
| - 93 Primary Health Care Centres |
| - Unity, Jonglei, and Upper Nile states, South Sudan |
| - 9,617 training observations, 2021 to 2024 |
| - Top facility: NHIAL DIU PHCC, Rubkona County, Unity State, CDI 0.501 |
|
|
| ## Contact |
|
|
| Makarere AI Research Lab |
| mubarakatanvebb88@gmail.com |
|
|
