A Novel Student Achievement Prediction Method Based on Deep Learning and Attention Mechanism

• Published in: IEEE access Vol. 11; pp. 87245 - 87255
• Main Authors: Liu, Yu, Hui, Yanchuan, Hou, Dongxu, Liu, Xiao
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 2023; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Achievement prediction; attention mechanism; characteristic selection; Correlation; Data mining; Deep learning; Education; Feature extraction; Model accuracy; Neural networks; Prediction models; Predictive models; Quality of education; Students
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2023.3305248

Student achievement is an important indicator for evaluating the quality of education. It can assess development potential of students and teaching level of lecturers. Predicting student achievement is an important aspect of education data mining, which can help teachers to guide learning process of students to improve student achievement and the quality of education. Existing methods for predicting achievement less focus on the correlation between influencing factors and student achievement, and ignore the influence of different factors on student achievement. Therefore, these models cannot achieve personalized analysis and guidance for students. To address these problems, this paper proposes a student achievement prediction model based on deep learning and attention mechanism (MCAG). The proposed model can study the correlation between various factors and highlight the influence of important factors. Firstly, the correlation between influencing factors and student achievement is analyzed using the maximum information coefficient and to determine the appropriate input parameter dimensions. Then, deep learning is used to extract high-dimensional and temporal features of the data, and the attention mechanism was used to effectively identify the importance of different attribute features for grades. Finally, the model predicts the final grades based on the fused features. The prediction performance of the proposed model has been validated through experiments, and compared with other baseline models, the accuracy of the proposed MCAG model is 94.22%, which indicates that the proposed model can predict student achievement more accurately.