Machine learning and reliable inferences

This working package will produce state-of-the-art machine learning models to be used in public health, but also a more general understanding of the process of model building. The underlying basic questions are the following: How do different inferential goals, like predictive modelling, causal inference, or algorithmic decision-making, influence the requirements set on the data? Second, how do the messiness and various possible deficiencies of register data – missing or unobservable data, sample distortion, limited variation, changing definitions, etc. – affect the uncertainty of the inferences that can be made from it? Third, what is the role of background knowledge and expert judgment in building and developing machine learning models?