An artificial intelligence driven adaptive learning system for enhancing student engagement and academic achievement in STEM education

Abstract

Students’ resistance to learning in Science, Technology, Engineering, and Mathematics (STEM) subjects, as well as academic disparities among students and insufficient individualized instructional support in STEM subjects, remain challenges for STEM education. The use of artificial intelligence (AI) to create adaptive learning systems emerges as a promising solution for anticipating learners’ needs and informing timely educational decision-making. This study proposes an adaptive learning system based on AI algorithms for predicting student engagement, forecasting academic performance, recommending personalized STEM learning strategies, and providing explainable decision support. The proposed framework has been tested with two publicly available educational datasets, one for engagement classification and achievement prediction, the Open University Learning Analytics Dataset (OULAD), and another for simulated STEM knowledge-tracing validation, ASSISTments 2009–2010. The model incorporates student features, learning-behavior patterns in time series, feature fusion via an attention module, and multiple-task learning branches for engagement classification and achievement regression. A rule-based recommendation module maps the model outputs to an adaptive support category, and an explainability analysis determines the most influential features for the prediction. Engagement labels were created as indicators derived from activity in the VLE, login frequency, assessment submissions, and forum participation. For behavioral engagement-proxy classification, the proposed model achieved an accuracy of 0.958 ± 0.004, an F1-score of 0.959 ± 0.004, and a macro-averaged one-vs-rest AUC of 0.997 ± 0.001 on OULAD. But XGBoost performed best overall in predictive performance, especially for predicting achievement. The simulated accuracy on ASSISTments was 0.824, the AUC was 0.843, and the F1 was 0.821 for the proposed model. The results show that the proposed framework can support behavioral engagement-proxy classification, achievement forecasting, STEM knowledge tracing, adaptive recommendation, and explainable education decision support. Causal intervention effects must be validated in the future in the live classroom.