Extracting topological features to identify at-risk students using machine learning and graph convolutional network models

Abstract Technological advances have significantly affected education, leading to the creation of online learning platforms such as virtual environments and massive open courses. While these offer a variety features, none them incorporates module that accurately predicts students’ academic performance commitment. Consequently, it is crucial design machine (ML) methods predict student identify at-risk students early possible. Graph representations data provide new insights into this area. This paper describes simple but highly accurate technique for converting tabulated graphs. We employ distance measures (Euclidean cosine) calculate similarities between construct graph. extract graph topological features ( GF ) enhance our data. allows us capture structural correlations among gain deeper than isolated analysis. The initial dataset DS can be used alone or jointly improve predictive power ML method. proposed method tested on an educational returns superior results. use compared with $$DS + GF$$ D S + G F in classification three classes: “failed”,“at risk”, “good”. area under receiver operating characteristic curve (AUC) reaches 0.948 using , 0.964 . accuracy case varies from 84.5 87.3%. Adding improves by 2.019% terms AUC 3.261% accuracy. Moreover, incorporating through convolutional network (GCN), prediction enhanced 0.5% 0.9% cosine matrix. With Euclidean matrix, adding GCN 3.7% 2.4%. By embedding models, identified 87.4% 0.97 AUC. solution provides tool detection students. will benefit universities their performance, improving both effectiveness reputation.