Robust Prediction when Features are Missing

TL;DR

This paper introduces a robust prediction method designed to handle missing features during prediction, leveraging the properties of an oracle predictor, and demonstrates its effectiveness on various datasets.

Contribution

It proposes a novel approach for robust prediction with missing features, grounded in oracle predictor properties, and validates its robustness through experiments.

Findings

Effective robustness to missing features demonstrated on real data

Approach outperforms standard methods in missing feature scenarios

Theoretical basis rooted in oracle predictor optimality

Abstract

Predictors are learned using past training data which may contain features that are unavailable at the time of prediction. We develop an approach that is robust against outlying missing features, based on the optimality properties of an oracle predictor which observes them. The robustness properties of the approach are demonstrated on both real and synthetic data.