International Journal of Sub-Saharan African Research (IJSSAR)

Background: Monitoring of viral load among pregnant and breastfeeding women augments remote patient management, reduces the risk of mother-to-child transmission of Human Immunodeficiency Virus (HIV), helps prevent treatment failure and virological rebound.
Objective: This study aimed to develop a machine learning (ML) model that effectively classifies the medical care status of HIV patients, particularly among pregnant and breastfeeding women, using integrated historic data of people living with HIV (PLHIV) in Oshana region, Namibia.
Method: A quantitative approach was employed to a cross-sectional dataset of 27,768 patients, from which 22,347 active patients were selected. Feature selection using a Random Forest classifier was used to reduce the risk of model overfitting. Three supervised learning models Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and a hybrid CNN-LSTM were trained using an 80/20 train-test split. Models were trained under two scenarios: (1) using all 71 demographic and clinical features and (2) using a reduced set of 5 top feature. Results: The hybrid CNN-LSTM achieved the highest performance (99.98% accuracy, 98.46% recall, 99.22% F1-score) and maintained strong results even with fewer features. In contrast, CNN and LSTM models showed reduced recall, highlighting the hybrid model’s superior ability to minimize false negatives, critical for identifying high-risk PBFW.
Conclusion: ML models can enhance healthcare decision making by providing accurate predictions to strengthen continuity of HIV care.
Unique Contribution: This study provides localized evidence on HIV care in Oshana region, Namibia by applying deep learning to classify the medical care status of pregnant and breastfeeding women. It demonstrates how routine clinical data can support scalable, data-driven interventions to improve continuity of care and reduce treatment failure in resource-limited settings. Key recommendation: Future research should explore alternative hybrid deep learning architectures, optimize complex hyperparameters, and evaluate diverse feature selection techniques. Testing on larger datasets is also recommended to assess scalability and generalizability.