Augmented Reality for STEM Education: A Scalable Model for Remote Classrooms

A remote STEM learning system based on augmented reality (AR) is introduced, which is scalable and has a structured multi-modal dataset and machine learning evaluation pipeline. The system gathers AR interaction logs, learner engagement, and device metadata along with learning assessment to sim-ulate student behavior and forecast learning outcomes. They are run through a standardized procedure of data construction such as anonymization, normaliza-tion of engagement, labeling at the concept level, and quality control. Performance measurement in the evaluation pipeline is in the form of the following metrics: confusion matrix, ROC curve, precision-recall, accuracy and loss convergence. The experimental outcomes provide the prediction accuracy of 97.4 and ROC-AUC of 0.986, which means high discriminating power and high generalization without overfitting. The results indicate that the patterns of AR interaction are robust predictors of conceptual mastery in remote STEM learning which offers a reproducible paradigm of scalable AR-based learning models as well as AI-driven custom education in the future.