Research Article

Human-Centered Risk Analytics for Household Malaria Stratification: An Explainable Decision-Support Framework

Abstract

Household-level malaria risk profiling is a high-impact decision-support problem because public health teams must allocate prevention, surveillance, and follow-up resources under uncertainty. This paper develops and evaluates a human-centered artificial intelligence and machine learning framework for classifying household malaria infection risk while preserving expert oversight. The study uses the supplied project outputs from a synthetic household dataset designed to emulate variables commonly collected in demographic and malaria indicator surveys, including housing materials, bed-net availability and use, proximity to water, household size, education, income, eave openness, flooring, window screening, and binary infection status. The pipeline imputes missing values, transforms skewed environmental distance, encodes ordinal and nominal attributes, balances the training data with synthetic minority over sampling, trains interpretable and ensemble classifiers, evaluates performance on a stratified hold-out set, and produces global and local explanations using feature-attribution visualizations. Logistic regression achieved the strongest overall accuracy (0.805), area under the receiver operating characteristic curve (0.87), and calibration score (Brier score 0.143), whereas the decision tree achieved the highest infected-household recall (0.756), indicating its usefulness when missed infections are operationally costly. Explainability outputs identified bed-net use, income level, proximity to water, flooring, bed-net availability, window screening, mud walls, household size, and thatch roofs as influential risk drivers. The paper contributes a reproducible AI/ML workflow, a human-inthe-loop review mechanism based on calibrated risk tiers and disagreement triggers, and an ethical governance framework for fairness, accountability, and deployment monitoring. The findings support the thesis that artificial intelligence should augment public health expertise rather than replace human judgment.