Bayesian network-based spatial predictive modelling reveals COVID-19 transmission dynamics in Eswatini

The first case of COVID-19 in Eswatini was reported March 2020, posing an unprecedented challenge to the country’s health and socio-economic systems. Using geographic information system (GIS) data comprising 15 socioeconomic, demographic environmental variables, we model spatial variability transmission risk based on for period under strict lockdown (up 8th May 2020) after regulations were gradually eased 30th June 2020). We implemented tested 13 data-driven Bayesian network (BN) learning algorithms examine factors that determine distribution risk. All BN models performed very well predicting cases as evidenced by low log loss (0.705–0.683) high recall values (0.821–0.836). tree-augmented naïve (TAN) outperformed all other algorithms. proximity major facilities, churches, shopping centres supermarkets average annual traffic density strongest predictors during lockdown. After gradual relaxation lockdown, proportion youth (15–40 years old) area became predictor addition areas where people congregate, excluding churches. study provides useful insights spatio-temporal dynamics drivers thereby aiding design geographically-targeted interventions. findings also point robustness predictive modelling graphically explaining phenomena uncertainty with limited data.