ENHANCING ANTENATAL CARE IN SOUTH AFRICA THROUGH URBAN DIGITAL TWIN TECHNOLOGY: INTEGRATING PM2.5 AND DEMOGRAPHIC DATA FOR IMPROVED MATERNAL AND NEONATAL OUTCOMES IN LMICS.

This project aimed to integrate urban digital twin technology with antenatal care programs to enhance maternal and neonatal health outcomes in South Africa, specifically focusing on the Gert Sibande District in Mpumalanga province. The study developed a predictive model for PM2.5 concentrations, using machine learning and geospatial data, and integrated these predictions into a digital twin framework for healthcare decision-making. The motivation for this work stems from the urgent need to mitigate the public health risks associated with air pollution, particularly PM2.5, which has been identified as a leading contributor to adverse health outcomes globally. In South Africa, where air pollution levels often exceed the World Health Organisation (WHO) guidelines, pregnant women face significant risks, including preterm birth and congenital anomalies such as orofacial cleft lip and palate [1]. Despite this, there has been limited integration of air quality data into healthcare frameworks [33] in low- and middle-income countries (LMICs). To address this challenge, this study used 2023 as the base year for training the predictive model due to computational requirements. The dataset included meteorological, chemical, and geospatial parameters, derived from ground-based air quality monitoring stations, satellite data, and demographic information. XGBoost, a robust machine learning algorithm, was applied with advanced hyperparameter optimisation techniques via Optuna to ensure high accuracy [33]. The final model achieved a cross-validation average root mean square error (RMSE) of 5.8004 and an R² score of 0.9004, indicating strong predictive performance. Key hyperparameters included a maximum depth of 12, a learning rate of 0.10086, and 680 estimators, among others. Feature importance analysis revealed that factors like boundary 1 layer height, wind patterns, and chemical pollutants such as NO₂ and SO₂ were critical in predicting PM2.5 levels. Interactive visualisations were implemented using Kepler.gl, showcasing pollution hotspots and temporal patterns. The WHO guidelines on PM2.5 were embedded into the digital twin framework to classify air quality into categories: green (safe), amber (moderate), and red (dangerous). The application provided real-time policy recommendations, such as advising pregnant women to stay indoors during high pollution events and wearing masks when outdoors. Future research included focusing on expanding the dataset beyond 2023 to improve temporal generalisability and predictive accuracy [20]. The integration of real-time air quality monitoring systems with healthcare alerts remains a key area for development, enabling dynamic, proactive interventions. Additional exploration of temporal forecasting models and pollutant interactions would further enhance the digital twin’s utility. Legal, social, and ethical considerations were addressed, including ensuring the use of anonymised and aggregated data to protect individual privacy. Ethical guidelines were followed to avoid misuse of personal health data, and all geospatial data complied with local regulations. This study demonstrated the feasibility and benefits of integrating digital twin technology with antenatal care programs to reduce air pollution exposure risks. The scalable framework developed here has the potential for adoption across LMICs, enabling targeted interventions to improve maternal and neonatal health outcomes.