Machine Learning-Based Prediction of Malaria Risk Among Rural Households in Kebbi State, Nigeria

Malaria remains a major public health burden in rural sub-Saharan Africa, where environmental and socioeconomic factors interact in complex, non-linear ways that traditional statistical models often fail to capture. This study examines machine learning-based prediction of malaria risk among 200 rural households in Kebbi State, Nigeria, integrating environmental exposure data (including proximity to stagnant water) with socioeconomic and behavioural survey indicators collected through a door-to-door household survey. Ensemble learning techniques, namely Random Forest and XGBoost, were applied alongside Decision Tree, Support Vector Machine, and logistic regression to identify the most significant predictors of malaria infection and to compare predictive performance across algorithms. Model performance was evaluated using accuracy, precision, recall, F1-score, and AUC-ROC on a held-out 30% test set following stratified 10-fold cross-validation hyperparameter tuning. Logistic regression and XGBoost achieved the strongest discrimination (AUC = 0.90 and 0.90, respectively), with Random Forest close behind (AUC = 0.89); consistent insecticide-treated net (ITN) use emerged as the single strongest predictor across all methods, followed by prior malaria history, stagnant water near the house, and overgrown bushes around the house. These findings indicate that machine learning approaches, including but not limited to ensemble methods, can support household-level malaria risk identification in resource-limited rural settings, offering actionable insight for targeted vector-control and behavioural interventions.