Environmental Determinants of Healthcare Demand: An Explainable ML Approach for Lisbon’s Air Quality-Health Nexus

Helder Relvas, Behrouz Nemati, Sina Ataee, Farzaneh Abedian Aval, Vânia Martins, Bárbara T. Silva, Diogo Lopes, Ana Isabel Miranda

Keywords: Environmental health, Healthcare demand prediction, Air pollution

TL;DR: Using 12 years of daily air quality, weather, and hospital data from Lisbon, this study employs explainable machine learning to uncover key environmental drivers of healthcare demand.

Abstract: This study applies machine learning (ML) to predict hospital admissions influenced by air pollution and meteorological conditions in Lisbon (Portugal), focusing on Hospital de Santa Maria. Four models, Artificial Neural Networks (ANNs), Random Forest, Extreme Gradient Boosting (XGBoost), and Histogram-Based Gradient Boosting Regressor (HGBR), were trained using air quality (PM2.5, PM10, NO2) and weather variables (temperature, humidity, pressure, wind). HGBR achieved the best performance (Tuning R2: 0.722, Testing R2: 0.521). SHapley Additive exPlanations (SHAP) analysis also showed temperature, particulate matter, and NO2 as key factors. The results highlight that combining gradient boosting with explainable AI provides a reliable, data-driven framework for forecasting hospital demand under changing environmental conditions.

Serve As Reviewer: ~Helder_Relvas1

Submission Number: 42