Removing Spurious Features Can Hurt Accuracy and Affect Groups Disproportionately

Explore a thought-provoking conference talk that delves into the unintended consequences of removing spurious features in machine learning models. Examine how this practice, often aimed at improving model performance, can paradoxically lead to decreased accuracy and disproportionately affect certain groups. Through a comprehensive analysis of various datasets and experimental setups, gain insights into the complex relationship between feature selection, model accuracy, and fairness in AI systems. Understand the implications of these findings for developing more robust and equitable machine learning algorithms, and consider the broader ethical considerations in AI research and development.

Introduction
Accuracy Drop
Setup
Data Sets
Results
Other Results
Conclusion