Fairness for machine learning software in education: A systematic mapping study

Abstract

The integration of machine learning (ML) systems into various sectors, notably education, has great potential to transform business workflows and decision-making processes. However, this technological advancement brings forth critical ethical concerns, particularly concerning the fairness of decisions affecting diverse groups of people. Our objective was to systematically map out the landscape of ML fairness research in higher education by exploring seven key research questions. These questions span a range of topics from the types of ML algorithms used in education to the methods of fairness assessment and the results achieved in terms of equity. We included 63 primary studies published between 2002 and 2023. The most common setting for AI Fairness research are: traditional machine learning algorithms (Logistic Regression, Random Forest, Decision Tree), sensitive variables (gender, race, ethnicity), and various definitions of fairness (Group fairness, Demographic parity, Equalized odds). We also identify several future research directions, including fairness assurance for multiple sensitive variables, combining different fairness concepts and metrics, open-source benchmarking tools, and fairness testing for modern ML/AI models.