This work aims to develop a fairness-aware model while maintaining competitive predictive accuracy in online learning environment. We retrieved a large dataset that included 3,145,728 video watch log entries and 893,189 assessment results from 14,252 students on an online learning platform. Six classic machine learning (ML) models were built to predict changes in student summative performance. Our findings indicate that prediction bias exists for groups with varying demographics when using models that lack fairness awareness. To enhance fairness, we optimized the existing treatment equality metric, which was previously limited to assessing a single sensitive attribute, to evaluate the overall fairness on multiple attributes of the prediction model. Additionally, we introduced a treatment equality loss function penalty term to constrain the models' training. The results demonstrate that our method can achieve comparable predictive performance while ensuring treatment equality across different groups.