Prediction of Medical Care Abandonment in Italy
a research project using machine learning to identify individuals most at risk of forgoing necessary healthcare services
This project investigates medical care abandonment in Italy using the 2019 European Health Interview Survey (EHIS).
The analysis combines machine learning models (multinomial logistic regression, XGBoost) with advanced techniques for feature selection, uncertainty quantification, and interpretability.
Key methods include:
- LASSO regression and gain-based importance for feature selection
- Bootstrap stability analysis
- Conformal prediction for uncertainty-aware classification
- SHAP analysis for model interpretability
Results show that XGBoost outperforms logistic regression, renunciation indicators (dental care, medication, mental health) are the strongest predictors, and conformal prediction provides reliable coverage. These findings highlight regional disparities and vulnerable population groups, offering actionable insights for policymakers.
Machine learning applied to healthcare abandonment in Italy