Abstract: In order to improve early intervention and support systems in educational settings, this study presents a multi-class adaptive active learning framework to predict student anxiety. Because of their static learning processes and lack of labeled data, traditional anxiety prediction models frequently perform poorly. Our method improves model accuracy and robustness by iteratively choosing the most informative data points for labeling using adaptive active learning. The model provides a nuanced understanding of students' mental health by distinguishing between different levels of anxiety through the use of multi-class classification. The effectiveness of the suggested approach is demonstrated by experimental results, which show notable gains in prediction accuracy over baseline models. With its scalable solution for real-time anxiety prediction and contribution to more responsive learning, this study highlights the potential of adaptive active learning in educational data mining.

Keywords: Adaptive Active Learning, Multi-ClassClassification, Student Anxiety Prediction, Educational Data Mining,Machine, Learning in Education,Mental ,Health Assessment,Real-Time Analytics.

| DOI: 10.17148/IARJSET.2025.125349