Methods and systems for robust supervised machine learning

Abstract

A disclosed method may include iterating a model optimization process, the iterating including one or more iterations. The method may also include updating a classification model based on the iterating, the updating performed using training data. The method may further include generating a final version of the classification model based on a final iteration. The method may also include setting a parameter (q), the parameter corresponding to a total number of observations (Q) that are to be removed from the training data by the final iteration. The method may further include determining one or more corresponding numbers of observations to remove from the training data, where the corresponding number of observations are to be removed at some of select iterations t, and the corresponding number of observations are to be removed based on the number Q and an estimate of the number of iterations remaining until the final iteration.