MikeDaMayne/malaria-forecasting-mozambique

Malaria incidence forecasting in Mopeia, Mozambique using epidemiological and climate data. Includes preprocessing, feature engineering (Malaria Proneness Index), and predictive models (Random Forest, Lasso, Ridge, ElasticNet, KNN, Decision Tree, OLS) # Malaria Incidence Prediction using Climate and Epidemiological Data This project investigates how lagged climate and epidemiological variables influence malaria incidence using machine learning models, which were trained by epidemiological data on Mozambique. ## Problem Statement Malaria remains a major public health challenge in many regions, particularly Africa. Accurate prediction of malaria incidence can help for planning and resource allocation for intervention programs in burdened regions. This project aims to model malaria incidence using climate data and epidemiological data, with a focus of assessing the importance of temporal (lagged) features and engineered features. ## Dataset Two datasets were used: - **Epidemiological dataset**: contains malaria incidence, testing rates, and distance from health facilities of some villages. - **Climate data**: contians rainfall, temperature, humidity, and other climatic data related to malaria incidences and outbreaks. All the climate variables were also lagged. Climate data was sourced from NASA POWER. The final dataset fed to models was made by integrated the two datasets. Important features include: - spray_satus - total_residents - clinic_tests - total_rdts - avg_distance_to_health_facility_km - positivity_rate - prev_month_incidence_per_1000 - rainfall - prev_month_rainfall - mpi (Malaria Proneness Index, Engineered feature) ## Methodology - Data was split 80/20 into training and testing tests chronologically for two reasons: - To preserve temporal structure - To avoid overfitting to a single village with generally high incidence. - Feature engineering included: - Lagged variables (by up to 4 months) for all climate variables and incidence. - A unique malaria proneness index (MPI) assigned to each village based on its general vulnerablility to malaria, accounting for the high spatial variability of average incidences in the data. Calculated using distance from health facilities, spray_status, lagged inci …