Question: This connects directly to the data preparation phase in ways that I think get overlooked. Before any fairness metric is even applied, decisions are being made about which variables to include, how to handle missing data, and what the target label actually represents. Each of those choices encodes assumptions. If a team is predicting "creditworthiness" using historical loan repayment data, and that historical data reflects decades of discriminatory lending, then cleaning and preparing that data without interrogating its origins means the bias gets laundered into the model quietly. The wrangling stage is where a lot of the most consequential value judgments happen, and it rarely gets the ethical scrutiny it deserves. D'Ignazio and Klein (2020) make a point that resonates here: data is never raw. It is always collected by someone, for some purpose, within some set of power relations. That framing helped shift how I think about preprocessing. It is not a neutral technical step. It is a site of decision-making that shapes what the model can even see. Need Assignment Help?