Fairness without Imputation: A Decision Tree Approach for Fair Prediction with Missing Values

Watch a research presentation from the 2022 Symposium on Foundations of Responsible Computing where Harvard University researcher Haewon Jeong explores fair machine learning approaches for handling missing data values. Learn about the challenges of applying fairness interventions when data missing patterns correlate with sensitive group attributes like gender or race. Discover a novel decision tree-based method that avoids explicit data imputation while optimizing for fairness through a regularized objective function. Follow along as the speaker analyzes different sources of discrimination risk in imputed datasets, presents theoretical foundations, and demonstrates experimental results showing how this integrated approach outperforms existing fairness intervention methods on real-world datasets with missing values. Gain insights into biased imputation, mismatched imputation methods, and future research directions for developing fair machine learning systems that can handle incomplete data responsibly.