Related Experiment Video
Updated: May 26, 2026
Visualizing Efficacy of Pesticides Against Disease Vector Mosquitoes in the Field
Published on: March 16, 2019
Applying Machine learning to analyze the utilization of insecticide-treated nets among rural under-five children in
Fentahun Bikale Kebede1, Amanuel Worku2, Angwach Abrham Asnake3
1Strategic Affairs Executive Office, Ministry of Health, Addis Ababa, Ethiopia.
Introduction:
Malaria continues to be a major cause of morbidity and mortality in children under the age of five in the sub-Saharan region of Africa. Despite being one of the pillars of prevention, the use of Insecticide-Treated Nets (ITNs) is still low in rural East Africa. This research aimed to forecast the use of ITNs by rural communities with children in this region using machine learning (ML) and identify the most predictive factors.
Methods:
We analyzed pooled data from the Demographic and Health Surveys (DHS) in 11 East African nations between 2011 and 2022. Six ML classification models, K-Nearest Neighbours, Support Vector Classifier (SVC), Random Forest, Multilayer Perceptron (MLP), Naïve Bayes, and Logistic Regression, were developed to predict ITN adoption. Model performance was evaluated using Accuracy, Precision, Recall, F1-Score, and ROC-AUC. Shapley Additive exPlanations (SHAP) analysis was used to improve model interpretability and to rank feature importance.
Result:
In 98,684 weighted household samples, the prevalence of general ITN use was 47.12%. The MLP model generalized best overall, achieving the highest accuracy (91.56%) and ROC-AUC (97.15%), with very high precision (87.95%) and recall (95.24%). While the SVC model had better recall (99.92%), its precision (84.19%) was significantly lower, indicating a higher proportion of false positives. MLP's F1-score of 91.39% restored it as the most balanced classifier in this scenario. SHAP analysis identified the number of children who slept under ITNs and the number of ITNs in the home as the strongest predictors. More community-level wealth, household wealth, and media exposure were also positively associated with ITN use.
Conclusion:
Machine learning made accurate predictions of ITN use, with the MLP model having the best and most balanced performance. SHAP allowed us to uncover actionable reasons for the most influential drivers of ITN use. These evidence-based outcomes can inform targeted public health interventions to improve ITN coverage and utilization in high-risk rural settings, ultimately reducing malaria cases and deaths.