Educational data analysis faces the challenge of optimizing predictive models for student performance. While traditional statistical and machine learning methods dominate, graph-based data representations remain underexplored. Graphs offer holistic insights into students’ learning journeys, revealing patterns beyond conventional models. The surge in educational data underscores the need to harness it effectively for student outcomes. This research Graph-Based Improvement of Student Performance Prediciton (GBISPP) bridges the gap by investigating graph-based methodologies for performance prediction. These techniques make relationships among students, courses, and resources, aiming to enhance predictive accuracy. In our study, After pre-processing and converting the data to graph. Then, we started by setting a threshold of (0.75), and we used (Gaussian filter) and (Spearman correlation) similarities for our research process. Then we added graph features (Cluster Coefficient, Betweenness Centrality, Eigenvector Centrality, Degree Centrality, Closeness Centrality, Average Weighted Degree, Average Clustering, Density, and Degree). The outcomes of our study for the five algorithms employed (Random Forest, Naïve Bayes, Decision Tree, AdaBoost, and SVM) are displayed, The Random Forest recorded the highest accuracy it was (86.34). For the data analysis process we used (5000) records in the Open University Learning Analytics Dataset (OULAD) from Kaggle, we focused on ‘studentInfo’ and ‘studentAssessment’ tables. This reliable source contributes to educational research in Educational Data Mining (EDM).
