Beyond Buffer-Based Approaches: Spatial Machine Learning for Urban COVID-19 Hotspot Prediction

This study identifies the most effective machine learning (ML) model for predicting COVID-19 hotspots by integrating four spatial indicators—facilities, mobility hubs, public spaces, and urban density—as proxies for social interaction intensity. While previous research has examined COVID-19 risks using buffer analysis, mobile-phone mobility data, and space-time models, this work is the first to systematically integrate these spatial indicators into a predictive ML framework designed for data-constrained urban contexts. Using zone-level case data from Sétif, Algeria, seven ML algorithms were evaluated, of which the Gaussian Process Regressor (GPR) achieved the highest predictive accuracy (R² = 0.97; RMSE = 14.11), outperforming traditional buffer analysis by approximately 15%. The framework demonstrates that combining multidimensional urban indicators captures nonlinear and interaction effects that shape transmission dynamics and yields policy-relevant insights for testing prioritisation, resource allocation, and density management. The model’s modular architecture enables recalibration for cities with differing infrastructure and mobility configurations, highlighting its adaptability across diverse urban environments.