Modelling Quality of Life among Adults in Gauteng Province, South Africa: A Machine Learning Approach

Background: Quality of Life (QoL) is a complex construct determined by a complex set of health, socio-economic and environmental determinants. Identifying the principal determinants of QoL and determinants that differ between subgroups of the population helps inform population-targeted intervention policies. This work sought to model QoL for adults in the Gauteng Province of South Africa using machine learning methods. Methods: The cross-sectional analytical study used the QoL Survey (QoL 2023–2024, Round 7) secondary data from the Gauteng City-Region Observatory with more than 14,000 adult resident responses. Five supervised machine algorithms—Logistic Regression, Random Forest, Support Vector Machine (SVM), Extreme Gradient Boosting (XGBoost), and Artificial Neural Networks (ANN), were trained and tested on accuracy, precision, recall, F1-score, and AUC-ROC. Results: Logistic Regression performed the best with 81% accuracy followed by SVM and XGBoost. In all the analyses the individual's health, perceived safety, work status appeared to be the most significant QoL predictors using the application of SHAP. Subgroup analysis results indicated model performance variability across levels of schooling, category of dwellings used etc. Prediction accuracy gender-wise differences remained not significant. Conclusion: Tlevel of satisfaction with the National Government, access to piped water, and toilet type possess the most influential effect on QoL in Gauteng Province. Logistic Regression is still a strong and interpretable QoL predicting model, with the assistance of the support of SHAP-based interpretation. The research suggests the necessity of health-oriented and place-oriented responses in policies to elevate the life quality of the South African cities' populations.