Deep Learning-Based Prediction of Student Performance in Physics Education Using Multimodal Data

This study proposes a deep learning-based approach to predict the academic performance of physics teacher trainees using multimodal data. By analyzing learning data in multiple dimensions such as classroom participation, homework completion, laboratory operations, online learning hours and discussion participation, we developed a prediction model to identify key factors affecting academic success. The results show that the deep learning approach can effectively capture the complex relationship between different learning modalities and final academic outcomes, and successfully identify key performance indicators in physics education while maintaining high prediction accuracy. The study found that online learning duration is the single most predictive indicator, while traditional classroom engagement and homework completion have relatively limited impact on learning outcomes, which provides new ideas and directions for improving physics education.