In causal inference [1], a central goal is to estimate the causal effect of a treatment variable \(T\) on an outcome variable \(Y\) . Observational data, where treatment assignment is not controlled, often suffers from confounding, leading to biased estimates of this effect [cite: 2, 3]. This chapter introduces two key criteria—the backdoor criterion [1] and the frontdoor criterion [1] which, under certain assumptions, allow us to identify causal effects from observational data. Backdoor and frontdoor criteria provide a causal framework for feature selection and model design in machine learning, ensuring that models learn genuine causal relationships. The Bias Problem: Machine learning models can learn biased relationships from observational data, leading to poor generalization and unfair predictions. Causal Solutions: The backdoor and frontdoor criteria provide principled ways to select features and structure models to mitigate this bias. Analogy: Model Debugging: These criteria are like debugging tools that help us identify and remove the sources of bias in our models.