In physics education, some students fail to have the foundational knowledge of energy concepts needed to engage in societal debates on climate change and energy transformation. A newly published study highlights the potential of AI to identify students with different learning trajectories and to help bridge the knowledge gaps.
The researchers used a digital workbook designed to teach energy concepts to collect detailed interaction data from over 500 students. After applying exclusion criteria, data from 172 students were analyzed to identify their productive and unproductive learning curves.
By using machine learning, specifically random forest models, and natural language processing (NLP), the researchers were able to classify students into productive and unproductive learning trajectories based on their post-test scores. The analysis revealed not only how students moved through the learning unit but also the types of knowledge they employed in open-ended responses.
The findings show how machine learning was able to differentiate between different learning trajectories and how NLP was able to give insights into the different knowledge elements that students used. These insights can help researchers and teachers to design more effective instructional materials and to assist in providing targeted feedback to teachers and learners.
For more information, check out the full article:
Wyrwich, T., Kubsch, M., Drachsler, H. & Neumann, K. (2025). Tracking students’ progression in developing understanding of energy using AI technologies. Physical Review Physics Education Research, 21 (010152). doi: https://doi.org/10.1103/PhysRevPhysEducRes.21.010152