Causal machine learning for reliable policy evaluation under complex real world conditions
Although predictive models are widespread, organizations still struggle to estimate what would happen after an intervention, especially with confounding, missing data, and changing populations. This project could develop robust causal machine learning methods combining structural causal models, representation learning, and sensitivity analysis. Novelty should target transportable treatment effects, hidden confounding, or policy optimization. Objectives include theoretical guarantees, benchmark datasets, transparent uncertainty estimates, and evaluation with domain experts and prospective interventions.