Fairness-Aware Streaming Feature Selection with Causal Graphs
Leizhen Zhang, Lusi Li, Di Wu, Sheng Chen, Yi He
TL;DR
The paper tackles fairness in streaming feature selection by modeling non-associational bias through two egocentric causal graphs centered on a protected attribute $S$ and the label $Y$. It introduces Streaming Feature Selection with Causal Fairness (SFCF), which incrementally constructs Markov blankets to identify strong, redundant, and irrelevant features and replaces inadmissible features with admissible ones (AD1/AD2) to balance empirical risk and sparsity under equalized odds constraints. The method jointly optimizes accuracy and fairness in an online setting, achieving superior fairness (lower EO) and sparsity with competitive accuracy across five benchmarks, while significantly reducing the number of selected features. The approach demonstrates scalable, real-time bias mitigation in streaming contexts and provides public code to support reproducible research. Overall, SFCF advances fair and efficient online feature selection by leveraging dual causal structures and conditional independence to adaptively manage bias in evolving feature streams.
Abstract
Its crux lies in the optimization of a tradeoff between accuracy and fairness of resultant models on the selected feature subset. The technical challenge of our setting is twofold: 1) streaming feature inputs, such that an informative feature may become obsolete or redundant for prediction if its information has been covered by other similar features that arrived prior to it, and 2) non-associational feature correlation, such that bias may be leaked from those seemingly admissible, non-protected features. To overcome this, we propose Streaming Feature Selection with Causal Fairness (SFCF) that builds two causal graphs egocentric to prediction label and protected feature, respectively, striving to model the complex correlation structure among streaming features, labels, and protected information. As such, bias can be eradicated from predictive modeling by removing those features being causally correlated with the protected feature yet independent to the labels. We theorize that the originally redundant features for prediction can later become admissible, when the learning accuracy is compromised by the large number of removed features (non-protected but can be used to reconstruct bias information). We benchmark SFCF\ on five datasets widely used in streaming feature research, and the results substantiate its performance superiority over six rival models in terms of efficiency and sparsity of feature selection and equalized odds of the resultant predictive models.