XGNN-AP: Explainable Graph Neural Network for Academic Performance

Abstract

Academic performance prediction has gained importance as a research domain in educational data mining, allowing for detection of at-risk students and taking proactive measures to support them. Traditional methods of machine learning have been focusing on treating the students as isolated individuals which makes it difficult for them to capture complicated relationships between the learner, course, teachers and learning environments. To overcome the mentioned drawbacks, we propose the framework XGNN-AP (Explainable Graph Neural Network for Academic Performance), a new explainable framework based on graphs which incorporates the Graph Attention Networks (GAT), Graph Convolutional Networks (GCN), and Explainable Artificial Intelligence (XAI) methods for precise and understandable prediction of academic performance. The framework consists of data pre-processing, heterogenous graph construction, graph representation learning, prediction generation, explainability via SHAP, GNNExplainer and graph attention visualization and finally, personalized recommendations and early warning modules. The proposed model has been tested according to common performance criteria such as accuracy, precision, recall and F1-score and then has been compared with traditional machine learning models such as Logistic Regression, Random Forest, Support Vector Machine, K-Nearest Neighbors, Multi-Layer Perceptron, Stacked Ensemble and Hybrid Neural Classifier.The experimental results proved the superiority of the suggested XGNN-AP model with the outstanding 99.71% classification accuracy, 1.00 macro precision, 0.99 macro recall, and 1.00 weighted F1-score compared to all the baseline models. At the same time, the application of SHAP-based explainability helped reveal the most crucial academic, behavioral, and demographic features influencing the students' performance, which ensured the interpretability and transparency of the predictions. Thus, the developed model can be considered a robust and interpretable decision support system for teachers.