This research explores the predictive modeling of acute morbidity related to air pollution, using spatiotemporal data and advanced machine learning techniques. The study focuses on respiratory tract infections and diseases with morbidity data sourced from Pirogov Hospital. Air quality data for particulate matter (PM2.5 and PM10) concentrations, were collected from five highly accurate monitoring stations and multiple lower-cost but spatially distributed sensors. The noise in the low-cost sensors’ measurements was mitigated using a two-step calibration method, ensuring data reliability. Three predictive models were implemented and compared: Long Short-Term Memory (LSTM), Random Forest (RF), and Support Vector Machine (SVM). LSTM demonstrated superior performance by capturing temporal dependencies, lag effects for different day windows, and cumulative exposure trends. The results reveal that predictive accuracy, significantly improved when spatial variations of air pollutants from multi-point monitoring stations were incorporated into the algorithm using a larger temporal window. These findings highlight the potential of combining deep learning and machine learning techniques with spatiotemporal air quality data to enhance healthcare systems by forecasting acute morbidity rates and enabling proactive resource allocation.