Machine Learning in Biomedical Informatics: Optimizing Resource Allocation and Energy Efficiency in Public Hospitals

This paper presents a computational framework that combines supervised machine learning and multi-objective optimization to support data-driven decision-making for resource allocation in public healthcare systems. The study uses real-world data from 51 public hospitals across Southern and Northern Italy, ensuring a representative sample of diverse healthcare structures. It analyzes the interplay between hospital admission volumes, adherence to quality standards, operational efficiency, and energy costs. The framework integrates several predictive models—including Random Forest, Support Vector Machines, and Logistic Regression—developed in Python using the scikit-learn library. Model performance was optimized using calibration techniques, while interpretability was enhanced through SHAP (SHapley Additive exPlanations), which identified the most influential clinical areas affecting both efficiency and energy expenditure. Feature importance results were used in a dual-objective linear optimization function to determine the optimal distribution of hospital admissions across nosological areas, balancing scale efficiency and energy cost reduction. Additionally, a multi-objective genetic algorithm (NSGA-II) was applied to optimize quality standard adherence levels, generating trade-off solutions along a Pareto frontier. The proposed approach offers a transparent, explainable, and scalable tool for healthcare administrators, providing actionable insights for strategic resource allocation. By integrating interpretable machine learning with computational optimization, the model contributes to building sustainable and high-performing hospital systems aligned with both operational and environmental objectives.