Supervised Learning with missing values: theoretical foundations and practical guidelines.

Missing values are ubiquitous in many fields such as health, business or social sciences. While much of the existing literature has concentrated on imputation and inference with incomplete data, supervised learning with missing values has received less attention. In this talk, I will discuss the challenges that missing values pose for regression and classification tasks. A common practical solution consists in imputing the missing values before learning. I will present the theoretical foundations of Impute-then-Regress methods and introduce a benchmark study addressing the question: do we need to impute accurately to achieve good predictions?