BadriAI-Hub/Dengue-Intelligent-Dashboard

LSTM-based dengue outbreak early-warning system for Khartoum State, Sudan — built for conflict-affected, data-scarce environments. # Dengue Intelligent Dashboard (DID) **An LSTM-based early warning system for dengue outbreak prediction in conflict-affected, data-scarce environments — built for Khartoum State, Sudan.** **🔴 Live Demo:** dengue-intelligent-dashboard.streamlit.app --- ## The Problem Dengue outbreaks in Sudan are difficult to forecast using conventional surveillance systems (EWARS, IDSR, DengueNet) because: - Conflict has disrupted routine case reporting and health infrastructure - Ground climate stations are sparse or non-functional - Displacement patterns significantly alter transmission risk but are rarely incorporated into existing models DID addresses these gaps by combining satellite climate data, displacement tracking, and epidemiological records into a single deep-learning forecasting pipeline designed to work *despite* missing or irregular data — a common reality in conflict settings. ## What Makes DID Different | Challenge in Conflict Settings | DID's Approach | |---|---| | Missing/irregular case data | MICE imputation + Kalman smoothing | | No reliable ground weather stations | NASA POWER, CHIRPS, and MODIS satellite data | | Displacement-driven transmission risk | UNHCR/IOM displacement tracking as a model feature | | Noisy, sparse sequences | LSTM Masking Layer to handle missing time steps directly | | Systemic reporting bias | Bayesian hierarchical correction | ## Screenshot ## Model & Methodology - **Architecture:** LSTM (Long Short-Term Memory) network with a masking layer for irregular/missing sequences - **Inputs:** Satellite climate data (rainfall, temperature, vegetation), flood extent, displacement data, geolocated case records - **Preprocessing:** Multiple Imputation by Chained Equations (MICE), Kalman smoothing for noisy time series - **Bias correction:** Bayesian hierarchical modeling to adjust for uneven reporting across regions - **Primary evaluation metric:** Recall ≥ 0.90 (in outbreak early-warning, missing a true outbreak is far costli …