Application in Student Performance Prediction Using Graph Regularization Nonnegative Matrix Factorization

Student performance prediction is an important educational problem in designing personalized teaching and learning strategies. Although matrix decomposition or neural network based algorithms have achieved great success in prediction applications, there are still many problems. Firstly, they are semantic and interpretive inadequacies. Secondly, these methods only focus on the performance information of students, ignoring the additional information such the learning background and courses. In this paper, a graph regularization nonnegative matrix factorization algorithm (GNMF) is proposed for the grade prediction of students in programming courses. Compared with classical matrix decomposition and neural network methods, non-negativity of GNMF means that students' achievement information can be decomposed into the sum of scores of multiple ability elements. Moreover, GNMF generates adjacency matrix graph based on questionnaire data related to student programming, which is integrated into NMF objective function as regularization term. Experiments on education datasets demonstrate that, compared with other methods, GNMF has smaller prediction error and higher prediction accuracy. The research result could facilitate personalized teaching in higher education.