The Propensity-Weighted Causal Learner (PWCL) for Efficient Machine Learning
Yixuan Ding, Youxin Zhu, Zimo Qi, Qianqian Wang, He Wang, Sijin Wu · 2024
Recently, much attention has been focused on the causal knowledge to enhance the predictors’ interpretability and performance in data science. This study proposes a novel propensity-weighted causal learner (PWCL) that effectively integrates causal information into the prediction by leveraging the data’s causal structure. The mathematical proofs and numerical simulation demonstrate the dual strengths of PWCL: the predictor keeps the accuracy of traditional machine learning approaches and achieves this in a fraction of the training time. In addition, a real-world scenario reveals that PWCL could potentially solve the Out-of-Distribution (OOD) problem of traditional machine learning. The findings highlight that propensity-weighted learning could be extended to a wide range of existing machine learning models to enhance performance while reducing the computational cost.