Machine learning algorithms application in WaSH prediction using DHS data in Sub-Saharan Africa

Abstract

Access to combined safe Water, Sanitation, and Hygiene services remains a major public health concern across Sub-Saharan Africa, with coverage disparities influenced by a wide range of individual and community-level factors. Traditional statistical analyses may not fully capture the complexity and interactions of these determinants. Therefore, the objective of this study was to identify predictors of WaSH in Sub-Saharan Africa based on DHS data using machine-learning algorithms. This study used regionally representative Demographic and Health Survey data from 33 Sub-Saharan African countries and applied machine learning techniques to estimate household-level access to combined water, sanitation, and hygiene (WaSH) services. This study was based on weighted data from 233,391 households, spanning survey years from 2010 to 2020. Important categorical predictors were chosen, including the household wealth index, age group, marital status, sex, household head’s educational attainment, and community-level variables. These variables were identified based on evidence from the literature, their theoretical relevance to WaSH outcomes, and their demonstrated statistical significance in exploratory and model-building analyses. The Synthetic Minority Over-sampling Technique (SMOTE) was used to address class imbalance after the data had been preprocessed and encoded. With Hyperparameter tweaking through grid search, seven machine learning models- Logistic Regression, Random Forest, Gradient Boosting, Decision Tree, k-nearest Neighbors, Naive Bayes, and Neural Network—were trained and assessed over the course of seven iterations. Accuracy, precision, recall, specificity, F1-score, and AUC were used to evaluate the model’s performance. The most significant predictors were found by extracting feature importance from the top-performing models. Combined WaSH service coverage was 19% among the included population in sub-Saharan Africa. Seven machine-learning algorithms were applied to predict access to WaSH services. The differences observed at the regional level are statistically significant. Gradient Boosting outperformed the other models in predictive performance ( p < 0.01). The most influential predictors of WaSH service access were the educational level of the household head ( p < 0.001), the household head’s age ( p < 0.05), and the household wealth index ( p < 0.001). Residence type ( p < 0.01) and media exposure ( p < 0.05) also made significant contributions. The analysis revealed that WaSH coverage was generally lower across the region, particularly in rural and less educated communities ( p < 0.01). In summary, machine-learning techniques helped identify key predictors of household WaSH (Water, Sanitation, and Hygiene) coverage, revealing that disparities were primarily associated with age, household wealth, and the educational level of the household head. These findings highlight the strong link between socioeconomic factors and access to essential WaSH services. Overall, WaSH coverage remained low across the dataset, highlighting persistent inequalities and the urgent need for targeted interventions.