MOOC Dropout Prediction Using Explainable Relational Graph Convolution

Résumé fourni par la source

This study addresses the issues of high rates of student attrition and the paucity of explainability of predictive models in Massive Open Online Courses (MOOCs) by proposing a framework for predicting student attrition based on an Explainable Relational Graph Convolutional Network (ERGCN). The model captures co-learning relationships among students and latent associations between courses by constructing a multidimensional relationship between students, courses, and behaviors. The proposed approach incorporates a dynamic temporal segmentation method with a 10-day cycle, in conjunction with a long short-term memory (LSTM) network, to extract fine-grained behavioral sequence features. The experimental results demonstrate that the validation based on the xuetangx data set indicates that ERGCN exhibits a remarkable predictive effect on the dropout ofMOOCstudents in higher education, with an F1 value of 92.56±0.04% and an AUC value of 88.84±0.03%. In comparison with the conventional benchmark model (logistic regression and support vector machine), the absolute enhancement of the AUC is 10.31-19.33 percentage points, and the relative enhancement is 13.1% -27.8%. The study utilizes an explainability module analysis to identify the key factors that influence dropout. The study provides MOOC platforms with a prediction tool that combines high accuracy and transparency. This facilitates the implementation of customized intervention strategies by educators for students who are at risk. For instance, they can prioritize the delivery of customized learning resources to the top 10% of students with the highest probability of dropping out. The present research proffers methodological references for relationship graph modeling and explainable artificial intelligence in the context of educational data mining.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.