HISP-Uganda/Acute-Malnutrition-Forecasting-Tool-AMFT-

# Acute Malnutrition Forecasting Tool (AMFT) AMFT is a Streamlit-based decision-support tool for district-level acute malnutrition monitoring in Uganda. It combines anomaly detection, short-term forecasting, geospatial visualization, data quality checks, and model evaluation to support early warning and operational planning. The current version focuses on three connected outputs: - 4-level anomaly classification for within-district and between-district monitoring - 3-level operational alerts derived from anomaly combinations - 3-month district forecasts of GAM caseloads with uncertainty bounds ## Core Capabilities - Load district monthly GAM caseload data from a local path or file upload - Load district boundaries from a local GeoJSON path or file upload - Classify observed GAM caseloads using district-relative and peer-relative anomaly rules - Generate operational alerts from within-district and between-district anomaly combinations - Forecast district GAM caseloads for the next 3 months - Display forecast uncertainty with 80% intervals - Compare districts or regions against a national reference trend - Explore Spearman correlations between GAM caseload and covariates - Review regression and classification model performance - Download classified observed data and forecast outputs ## Risk Framework ### 1. Anomaly Classification Observed and forecast anomaly labels use percentile thresholds: | Level | Threshold | Meaning | |---|---|---| | `Low` | Below 75th percentile | Within the usual range | | `Moderate` | 75th to below 90th percentile | Higher than usual | | `High` | 90th to below 95th percentile | Unusually high | | `Extreme` | 95th percentile and above | Exceptionally high | ### 2. Operational Alert Operational alerts are derived from the combination of within-district and between-district anomaly levels: | Operational Alert | Rule | Typical Action | |---|---|---| | `Monitor` | Within = `Low` and Between = `Low` or `Moderate` | Routine monitoring | …