Predicting academic performance of students in Chinese-foreign cooperation in running schools with graph convolutional network

• Published in: Neural computing & applications 2021-01, Vol.33 (2), p.637-645
• Main Authors: Hai-tao, Pu, Ming-qu, Fan, Hong-bin, Zhang, Bi-zhen, You, Jin-jiao, Lin, Chun-fang, Liu, Yan-ze, Zhao, Rui, Song
• Format: Article
• Language: English
• Subjects: Academic achievement; Artificial Intelligence; Artificial neural networks; Colleges & universities; Computational Biology/Bioinformatics; Computational Science and Engineering; Computer Science; Cooperation; Correlation coefficients; Data Mining and Knowledge Discovery; Image Processing and Computer Vision; Performance prediction; Probability and Statistics in Computer Science; S.I. : DPTA Conference 2019; Students; Support vector machines; System effectiveness
• ISSN: 0941-0643; 1433-3058
• DOI: 10.1007/s00521-020-05045-9

To improve students' performance effectively, many Chinese universities are establishing systems to predict student's academic performance and sending student's academic early alerts. With student's population of over 600,000, Chinese-foreign cooperation in running schools (CFCRS) has become one of the booming higher education forms in China. Compared with students in the non-cooperatively running programs, CFCRS students' academic performances are weaker on average. To predict the academic-at-risk students and provide efficient supports to the students, a precise and prompt academic prediction is in great need. Therefore, this research aims at representing a more efficient and accurate model to predict academic performance of CFCRS students which will be based on graph convolutional network. In this research, student's similarity is measured on academic performance by Pearson correlation coefficient. An undirected graph in which similar students are connected is conducted. Feature matrix is composed of students' previous grades. In addition, graph convolutional network is trained based on the undirected graph and feature matrix. In the experiment, it shows that this model predicts certain student's performance on certain course from the final exam results in previous semesters, which might improve learning efficiency and teaching quality. With an average accuracy of 81.5%, graph convolutional network outperforms support vector machine and random forest models.