Towards Fair Machine Learning Software: Understanding and Addressing Model Bias Through Counterfactual Thinking

Researchers developed a novel counterfactual approach to address fairness bugs in machine learning software that maintains competitive performance while improving fairness. The method outperformed existing solutions in 84.6% of cases across extensive testing on 8 real-world datasets using multiple performance and fairness metrics.

• →A new counterfactual thinking approach tackles root causes of bias in ML software without sacrificing performance.
• →The method combines models optimized for both performance and fairness to achieve optimal results in both aspects.
• →Extensive evaluation covered 10 benchmark tasks using 5 performance metrics, 3 fairness metrics, and 15 measurement scenarios.
• →The approach outperformed state-of-the-art solutions in 84.6% of overall cases across 8 real-world datasets.
• →This research addresses growing concerns about unfair and unethical decisions from increasing ML software deployment.