Abstract
Achieving human immunodeficiency virus (HIV) viral suppression is crucial for enhancing health
outcomes and mitigating transmission rates among individuals receiving antiretroviral therapy
(ART). This dissertation investigates the integration of predictive modeling and explainable
artificial intelligence (XAI) to improve the understanding and prediction of ART success among
people living with HIV (PLWH) in Uganda. This research aimed to develop, evaluate, and
interpret various machine learning (ML) models for predicting viral suppression, embedding
XAI techniques to enhance both the accuracy of predictions and the interpretability of the
underlying factors influencing treatment outcomes. A robust methodological framework was
employed, encompassing a retrospective analysis of clinical and demographic data from 1,101
Ugandan PLWH on ART . By utilising various ML algorithms—including logistic regression
(LR), stacked ensemble (SE), random forest (RF), support vector machines (SVM), extreme gradient
boosting (XGBoost) , k-nearest neighbours (KNN), na¨ıve Bayes (NB) and artificial neural
networks (ANN)—the study systematically compared modelperformance while addressing class
imbalance through Synthetic Minority Over-sampling Technique (SMOTE). XAI methodologies,
including SHapley Additive exPlanations (SHAP) and individual conditional expectation
(ICE) plots, were utilised to identify the predictors of viral non-suppression.The findings revealed
significant insights into the performance of different ML algorithms, with the XGBoost
model demonstrating superior accuracy (0.89) and specificity (0.94). The integration of XAI
facilitated the identification of critical factors affecting viral suppression, notably adherence to
treatment, World Health Organization (WHO) clinical stage, and patient support dynamics.
These insights not only enhanced model performance but also improved the transparency of the
decision-making process, enhancing trust among clinicians and patients in data-driven HIV care
strategies. This dissertation significantly contributed to the integration of predictive modeling,
XAI, and HIV treatment outcomes. It offered practical applications in personalised healthcare
iv
Abstract
interventions, particularly in resource-constrained settings. The work highlighted the relationship
between ML insights and clinical practice and advocated the incorporation of predictive
analytics in HIV care. This approach reinforced the transformative potential of machine learning
and XAI in addressing public health challenges. Future research directions may include
extending these methodologies to other chronic health conditions and integrating them in real
time into clinical decision-support systems. The ultimate goal is to optimise patient outcomes
and enhance public health interventions on a larger scale.