ChristianIshimwe7/PREDICTING-MALARIA-RISK-IN-RWANDA-USING-MACHINE-LEARNING

PROJECT PLAN: PREDICTING MALARIA RISK IN RWANDA USING MACHINE LEARNING. AUTHOR: Christian ISHIMWE/ African Leadership University Rwanda. E_MAIL: c.ishimwe7@alustudent.com Problem: Predicting malaria risk (high/low) in Rwandan communities using historical health and environmental data to support early interventions by the Rwanda Biomedical Centre (RBC). Why It Matters: Malaria remains a leading cause of morbidity in Rwanda, with 1.7M cases in 2022 (WHO). Predictive models can help target resources (e.g., bed nets, sprays) to high-risk areas, especially in rural regions. Mission Alignment: This directly supports your goal of using ML to improve healthcare access and outcomes in Rwanda. Dataset Source: Combine Rwanda Health Management Information System (HMIS) data (available via statistics.gov.rw) with open environmental datasets (e.g., rainfall, temperature from NASA POWER). HMIS: Provides malaria case counts, patient demographics, and clinic locations. NASA POWER: Offers weather data (rainfall, temperature) correlated with malaria spread. Backup Option: If HMIS access is limited, use WHO Global Health Observatory or Kaggle malaria datasets. Features: Rainfall, temperature, humidity, population density, historical malaria cases, age, gender. Target: Binary classification (high/low malaria risk). Justification: Relevant to Rwanda’s public health needs, rich for feature engineering, and not from sklearn/keras (meets assignment rules). Methodology Traditional ML (Scikit-learn): Models: Logistic Regression, Random Forest, XGBoost. Feature Engineering: Normalize weather data, encode categorical variables (e.g., province), create interaction terms (e.g., rainfall × temperature). Deep Learning (TensorFlow): Sequential API: Simple feedforward neural network (FNN). Functional API: Multi-input model combining weather and demographic data. tf.data API: Efficient data pipeline for batching and shuffling. Experiments: Vary hyperparameters (e.g., learning rate, nu …