<p>Online learning has become a fundamental component of higher education. Predictive analytics tools are increasingly used in online learning environments to provide performance insights and learning support. These analytical approaches continue to be evaluated for their effectiveness and scalability in diverse educational settings. Despite their potential benefits, concerns exist regarding equitable outcomes across different demographic groups and individual learners. Current research on AI fairness in education has established theoretical foundations while also highlighting some limitations. There are preliminary studies categorizing sensitive individual attributes into binary groups (e.g., economic status), which may not capture the complexity of real-world scenarios. Furthermore, defining and operationalizing individual fairness is potentially more challenging than group fairness, leading most research to focus only on the latter. However, it is crucial to ensure that AI algorithms provide similar individuals with similar predictions. To address these concerns, we propose a fairness-aware model called Con-LSTM. This model incorporates contrastive learning to promote fair feature learning for multi-level prediction tasks. We evaluated the group fairness (measured by Subgroup AUC) and individual fairness (measured by similarity-based consistency score) of Con-LSTM using the Open University Learning Analytics Dataset, comparing it against fairness-unaware LSTM model. Our findings show that Con-LSTM achieves both high predictive accuracy and fairness. By balancing these two critical aspects, our approach highlights the potential to offer fair and reliable insights into student performance. This advancement could lead to more personalized and equitable educational experiences, demonstrating the significant benefits of integrating fairness-aware principles into AI models used in education.</p>