Risk stratification aims to identify patients who need monitoring and follow-up and to reduce health costs. For that purpose, we use patient databases and state-of-the-art machine learning and deep learning approaches to predict health outcomes (readmissions, complications, deaths) and healthcare costs. Machine learning algorithms are designed to detect complex patterns and interactions. In administrative datasets, the number of observations often exceeds the number of variables, so there is no identification problem, statistically speaking. We apply costing methods to establish the value of certain interventions, medical procedures or investments in health.