Predicting HIV Status Using Machine Learning Techniques and Bio-Behavioural Data from the Zimbabwe Population-Based HIV Impact Assessment (ZIMPHIA15-16)

HIV and AIDS continue to be a significant public health concern globally, with about 36 million people currently living with the epidemic. Several HIV interventions have been implemented to intensify virus transmission prevention, screening, and diagnosis in sub-Saharan African countries, including Zimbabwe. HIV prevalence is substantially high in Zimbabwe despite the significant progress made in the previous years. As the country moves closer to attaining the epidemic control status, there is a need for targeted HIV interventions focusing on HIV risk individuals. Most current HIV interventions are based on evidence about specific sub-population groups, undermining the diversity of individual risk levels within such groups. Therefore, this study applied random forest classifier, support vector machine, and logistic regression to predict HIV status outcomes using Zimbabwe Population-Based HIV Impact Assessment data to identify high-risk individuals and develop targeted interventions based on risk. This study shows that logistic regression outperformed the random forest classifier and support vector machine with the prediction accuracy of 85%, recall of 98%, and F1-score of 92%. However, the random forest classifier has the highest precision of 87% compared to the other models. The support vector machine outperformed the random forest classifier in recall and F1-score metrics, with a recall of 96% and F1-score of 91%. Machine learning models can help identify individuals at high risk of contracting HIV and assist policymakers in developing targeted HIV prevention and screening strategies informed with socio-demographic and risk behavioural data. However, this study only used socio-demographic and behavioural predictors to predict HIV status. There is a need to include other HIV clinical predictors to optimise HIV status prediction models better and further integrate them into real-world healthcare settings.