Date
Publisher
arXiv
Fairness in artificial intelligence and machine learning (AI/ML) models is becoming critically important, especially as decisions made by these systems impact diverse groups. In education, a vital sector for all countries, the widespread application of AI/ML systems raises specific concerns regarding fairness. Current research predominantly focuses on fairness for individual sensitive features, which limits the comprehensiveness of fairness assessments. This paper introduces FAIREDU, a novel and effective method designed to improve fairness across multiple sensitive features. Through extensive experiments, we evaluate FAIREDU effectiveness in enhancing fairness without compromising model performance. The results demonstrate that FAIREDU addresses intersectionality across features such as gender, race, age, and other sensitive features, outperforming state-of-the-art methods with minimal effect on model accuracy. The paper also explores potential future research directions to enhance further the method robustness and applicability to various machine-learning models and datasets.
Study at a glance
- Question: Can a regression-based data debugging method improve fairness across multiple sensitive attributes simultaneously without harming model accuracy?
- Setting and data: Seven datasets: six from Kaggle (Adult, COMPAS, Default, Student Dropout, Student Performance, OULAD) and one from Dai Nam University in Vietnam.
- Method: Model evaluation testing a multivariate regression preprocessing method (FAIREDU) across Logistic Regression, Decision Tree, and Random Forest algorithms.
Why it matters
Educational machine learning models risk bias against historically marginalized groups. Existing fairness methods predominantly target single sensitive features, leaving models unable to handle the intersection of multiple sensitive factors like gender, race, and disability.
Key findings
- FAIREDU achieved 35 wins out of 60 fairness comparisons against the prior benchmark LTDD, reducing disparate impact by up to 96.7%.
- Model performance remained balanced after debiasing, with accuracy matching or exceeding baselines in 22 out of 45 cases and experiencing minimal drops overall.
- Bias varied across machine learning models and metrics, with Decision Trees demonstrating the lowest bias while Logistic Regression showed higher bias.
Summary drafted with AI from the full paper and may contain errors. See the full paper for details.
What is the application?
Who is the user?
Who age?
Why use AI?
Study design
