ekiru111/Uganda-Malaria-Spatio-temporal-ML

ML-based spatio-temporal analysis of malaria risk factors in Uganda integrating survey, household, geographic, and environmental data. # Predictive Modeling and Spatio-Temporal Simulation of Malaria Outbreaks in Uganda ## Author **Name:** Ekiru Ernest Ochepa **Program:** Kujenga AI/ML Course --- ## Objective ### **Overall Objective** This research develops a predictive machine learning model to identify malaria risk factors and forecast malaria outbreaks across Uganda. The project integrates multiple data sources including demographic surveys, household characteristics, geographic information, and satellite-derived environmental variables to build a comprehensive spatio-temporal analysis of malaria transmission patterns. ### **Specific Objective** 1. Develop a predictive machine learning model to identify malaria risk factors 2. Forecast malaria outbreaks across Uganda 3. Build a comprehensive spatio-temporal analysis of malaria transmission patterns --- # Results --- ## **Model Development and Performance** Two models were evaluated that is, Random Forest and XGBoost to determine the most effective approach for identifying positive malaria cases. While both models were tested, XGBoost demonstrated superior potential for clinical application due to its high sensitivity ***Baseline Performance (Positive Class: those with Malaria)*** - Random Forest: Recall 0.52 / Precision 0.30 - XGBoost: Recall 0.93 / Precision 0.19 ### **The Precision-Recall Trade-off** In malaria screening, the cost of a False Negative (missing an infected patient) is significantly higher than a False Positive (unnecessary follow-up testing). However, a precision of 0.19 is too low for efficient resource allocation. The goal here is to identify the "Sweet Spot" using the Precision-Recall Curve. The aim is to find an Optimal Threshold that maintains high diagnostic coverage while reducing "noise." - ***Target Metric:*** Maintain Recall at approximately 0.75 or above while optimizing Precision to approximately 0.25 and above. ### **Finding the "Elbow"** By plotting Precision against Recall, we can visually identify t …