Causal Regularization

TL;DR

This paper introduces a causal regularizer that improves the causal interpretability of predictive models, especially in healthcare, by enhancing causal accuracy and multivariate causation detection while maintaining competitive predictive performance.

Contribution

It proposes a novel causal regularizer, studies its theoretical properties, and demonstrates its effectiveness in healthcare data and neural network architectures.

Findings

Outperforms L1 regularization in causal accuracy on EHR data.

Enables non-linear causality analysis with neural networks.

Achieves up to 20% improvement in multivariate causation detection.

Abstract

In application domains such as healthcare, we want accurate predictive models that are also causally interpretable. In pursuit of such models, we propose a causal regularizer to steer predictive models towards causally-interpretable solutions and theoretically study its properties. In a large-scale analysis of Electronic Health Records (EHR), our causally-regularized model outperforms its L1-regularized counterpart in causal accuracy and is competitive in predictive performance. We perform non-linear causality analysis by causally regularizing a special neural network architecture. We also show that the proposed causal regularizer can be used together with neural representation learning algorithms to yield up to 20% improvement over multilayer perceptron in detecting multivariate causation, a situation common in healthcare, where many causal factors should occur simultaneously to have an effect on the target variable.