From Social to Academic: Associations and Predictions Between Different Types of Peer Relationships and Academic Performance Among College Students

This study aims to expose the correlation between different types of social behaviors and the academic performance of college students, and then to predict the academic performance of college students based on their social characteristics. We extracted and computed information on social relationships for roommates, classmates, members of the opposite sex, and others, based on real consumption data of 5597 freshmen students. The correlations between different types of peer relationships and academic performance were compared. Subsequently, we used Random Forests and Neural Networks as baseline methods, and introduced Graph Convolutional Network and Dynamic Graph Convolutional Network algorithms, on top of a graph network model based on social characteristics, to predict students’ academic performances. The results show that the quantity and quality of all types of socialization are positively correlated with academic performance, and socialization among classmates and roommates demonstrates a stronger correlation. In addition, with the construction of the graph model and the integration of time-series information, the prediction accuracy of the dynamic graph convolution method improved compared to other methods. The findings demonstrate the advantages of using social characteristics for academic performance prediction, and reveal the significant potential of AI applications in supporting the field of school management.