Integration of Causal Inference with Machine Learning for Improved Treatment Effect Estimation in Observational Studies

Abstract

Estimating treatment effects accurately in observational studies is a persistent challenge due to confounding, selection bias, and the non-random assignment of treatments. Traditional causal inference frameworks, while statistically grounded, often struggle with high-dimensional data and non-linear relationships. Conversely, machine learning (ML) excels in handling such data complexities but lacks a principled approach to causal interpretation. This paper explores the integration of causal inference techniques with machine learning models to improve the estimation of average and heterogeneous treatment effects (ATE, HTE) in observational studies. We discuss existing approaches, such as doubly robust learners, causal forests, and targeted maximum likelihood estimation (TMLE), and propose a synthesis framework grounded in the Neyman-Rubin causal model. Our results highlight that hybrid models significantly outperform traditional estimators under varied confounding scenarios and offer better generalizability in real-world applications.