What is causal inference evaluation and why is it different from predictive evaluation?

Causal evaluation measures effect of interventions, while predictive evaluation measures correlation-based accuracy.

Predictive evaluation focuses on how well a model predicts outcomes (P(Y|X)), using metrics like accuracy or RMSE. Causal inference evaluates the effect of interventions (P(Y|do(X))). The key difference is that causal evaluation requires counterfactual reasoning, controlling for confounders, and estimating treatment effects. Techniques include propensity scoring, instrumental variables, and randomized controlled trials (RCTs). Without causal structure, predictive models may perform well but fail under interventions.