Missingness Shift

Missingness shift, the change in the pattern of missing data between training and deployment datasets, poses a significant challenge for machine learning models. Current research focuses on developing robust prediction methods that account for various missingness mechanisms, ranging from ignorable (missingness independent of observed and unobserved data) to non-ignorable (missingness dependent on unobserved data), and exploring how adversarial manipulation of missing data can bias model learning. This research is crucial for improving the reliability and fairness of machine learning models in real-world applications where incomplete data is common, particularly in causal inference and domain adaptation tasks. Addressing missingness shift is essential for building more robust and trustworthy AI systems.