Non-imputational method for handling missing values in a prediction context, meaning that not only are there missing values in the training dataset, but also some values may be missing in future cases to be predicted. Based on the notion of regression averaging (Matloff (2017, ISBN: 9781498710916)).