Beyond Prediction Performance - How Modeling Decisions Shape Fairness Outcomes in Statistical Profiling

Explore how seemingly minor modeling decisions in machine learning systems can have profound consequences for fairness and equity when bridging from prediction to intervention in this 28-minute conference talk. Examine a comparative analysis using German administrative labor market data that reveals how different regression and machine learning approaches for predicting long-term unemployment risk achieve comparable predictive accuracy (ROC-AUC: 0.70-0.77) yet show striking disagreement in which individuals are classified as high-risk, with Jaccard similarities as low as 0.45 between equally accurate models. Discover how these differences cascade through the intervention pipeline, where classification thresholds, feature importance patterns, and model architectures each reshape the demographic and socioeconomic profile of those targeted for support. Learn about the critical challenge at the prediction-intervention interface where training data sufficient for forecasting outcomes still requires researchers to make consequential choices about which predictive voice to amplify. Understand the implications for documentation, transparency, and fairness in algorithmic decision-making systems, and consider how the prediction-to-intervention pipeline demands richer evaluation frameworks that account for both accuracy and equity in resource allocation.