Using Supervised Contrastive Learning for Improving Individual Fairness in AI Online Performance Prediction

• In Event: Data-Driven Insights in Online Learning

Abstract

Although there have been more studies focused on AI fairness in education in recent years, some are theoretical discussions and fairness evaluations. For those actively reducing bias and ensuring fairness in education, they often treat sensitive attributes as binary and focuses on group fairness. This study focuses on individual fairness and introduces a fairness-aware model using contrastive learning for fair feature learning in multi-level predictions. We compared Con-LSTM with fairness-unaware models on the Open University Learning Analytics Dataset. Results show that Con-LSTM ensures comparable predictive accuracy while improving individual fairness. Our approach highlights the potential for equitable AI-driven insights in education, fostering personalized and fair student experiences.

Authors

• Zifeng Liu, University of Florida

• Yihan Jiang, University of Florida

• Chenglu Li, University of Utah

• Wanli Xing, University of Florida