PRISM: Predicting Student Performance using Integrated Similarity Modeling with Graph Convolutional Networks

The widespread availability of educational data and technology-enhanced learning platforms has opened up new opportunities for leveraging learning analytics to address student issues, optimize educational environments, and enable datadriven decision making. Accurately predicting student performance is crucial for addressing key challenges in education, such as reducing dropout rates, facilitating personalized learning, and enhancing teaching efficiency. However, conventional methods often overlook the underlying relationships between students, limiting the effectiveness of performance prediction models. Thus, the proposed method, Predicting Student Performance using Integrated Similarity Modeling (PRISM), utilizes graph structures and Graph Convolutional Networks (GCN) to improve student performance prediction. In PRISM, a graph is constructed using academic performance and personal characteristics data, with each student represented as a node and edges denoting their relational similarities. GCN is applied to learn from this graph structure and predict students– academic performance. The proposed pipeline is evaluated using a real-world educational dataset and compared with traditional machine-learning approaches. Experimental results demonstrate the superior performance of the proposed GCN-based model in student performance prediction.