Student Performance and Prediction in Student Management Based on the Louvain Algorithm and a Hybrid Neural Network Model

Abstract

The efficiency and accuracy of existing student performance prediction technologies are low. This study proposes a prediction model that combines the Louvain algorithm and a hybrid neural network model. The model aims to improve the accuracy and efficiency of student performance prediction. The model uses graph folding technology to optimize the ILouvain algorithm and mines multi-level student community structures. The model also integrates the Transformer mechanism and the Channel Attention mechanism to extract multi-source features and strengthen key discriminative information. On this basis, the model combines the Kernel-K Nearest Neighbor to introduce a local similarity mechanism. Finally, the study builds a student performance prediction model. The loss value in the student performance prediction task in the proposed model is only 0.005. The coefficient of determination is 0.951. The proposed model maintains high training accuracy and test accuracy. The proposed model correctly clusters the student performance into four groups based on the grade point average, and the value of the silhouette coefficient is 0.9177. The proposed model maintains the prediction error of the scores. The Root Mean Square Error and mean absolute error are 1.80 points and 1.45 points. The research model has superior predictive performance in different datasets. Furthermore, in the previous semester, the reasoning time of the proposed model is 38.65[Formula: see text]ms. In the next semester, the average reasoning time is only 32.87[Formula: see text]ms. The proposed model accurately predicts student performance and shows strong generalization ability and inference efficiency. This study helps build accurate student performance prediction models in the future and provides technical support for the field of student management.