Predictive Analytics for Emergency Department Visits Based on Local Short-Term Pollution and Weather Exposure

Proper management of Emergency Rooms is needed to improve healthcare and patient satisfaction. Predicting accesses and hospitalisation rates through Machine Learning approaches appears promising, especially when coupled with air pollution and weather data. This work applies both Random Forest and AutoRegressive Integrated Moving Average approaches on data related to Brescia's clinical and environmental data from 2018 to 2022 to predict daily accesses or daily hospitalisations for cardiovascular and respiratory disorders. The predictions adhere quite well to the actual data for Random Forest, but less for AutoRegressive Integrated Moving Average. However, even if the specific value is not always correctly predicted, the overall trend seems to be rightly forecasted and performance metrics are mostly satisfying. Although additional work is required to improve their performances, results are encouraging and this sort of geographically-localised time-series forecasting seems feasible. Future developments will take into consideration the whole province of Brescia.