Controlling for Unobserved Confounds in Classification Using Correlational Constraints

  • 2018-01-11 15:35:43
  • Virgile Landeiro, Aron Culotta
  • 0

Abstract

As statistical classifiers become integrated into real-world applications, itis important to consider not only their accuracy but also their robustness tochanges in the data distribution. In this paper, we consider the case wherethere is an unobserved confounding variable $z$ that influences both thefeatures $\mathbf{x}$ and the class variable $y$. When the influence of $z$changes from training to testing data, we find that the classifier accuracy candegrade rapidly. In our approach, we assume that we can predict the value of$z$ at training time with some error. The prediction for $z$ is then fed toPearl's back-door adjustment to build our model. Because of the attenuationbias caused by measurement error in $z$, standard approaches to controlling for$z$ are ineffective. In response, we propose a method to properly control forthe influence of $z$ by first estimating its relationship with the classvariable $y$, then updating predictions for $z$ to match that estimatedrelationship. By adjusting the influence of $z$, we show that we can build amodel that exceeds competing baselines on accuracy as well as on robustnessover a range of confounding relationships.

 

Introduction (beta)

None

 

Conclusion (beta)

None