Modeling Student–Project–Skill Interactions Using Graph Neural Networks for Enhanced Learning Analytics

Abstract

Learning analytics is increasingly used to understand learner behavior, predict academic performance, and support personalized interventions. In contrast, traditional methods of machine learning algorithms assume that the learner is an isolated observation without considering the relational structure of educational scenarios. To overcome this problem, this paper presents an innovative solution in the form of a Student-Activity-Skill Graph Neural Network (SAS-GNN) model that captures the relationship between the learner, activity, and skill by representing them with a heterogeneous graph. SAS-GNN combines graph convolution, relation-aware message passing, graph attention mechanism, and the GNNExplainer technique to provide solutions to the task of predicting the outcomes of learning and interventions. When evaluated on the Open University Learning Analytics Data Set (OULAD) with more than 10 million interactions of learners, SAS-GNN attained 92.4%, 0.915, and 0.946 accuracies, F1-scores, and ROC-AUC scores, respectively. The results highlight the effectiveness of graph-based learning analytics in delivering accurate, interpretable, and actionable educational insights.