BISAP: A student academic performance prediction model based on the fusion of classroom behavior images and educational information

Student academic performance prediction is a significant area of study in the realm of education that has drawn the interest and investigation of numerous scholars. The current approaches for student academic performance prediction mainly rely on the educational information provided by educational system, ignoring the information on students’ classroom behaviors, which has an impact on the model’s prediction accuracy. To address this issue, this study proposes a student academic performance prediction model based on the fusion of students’ classroom behavior images and educational information (BISAP). This model aims to integrate students’ classroom behavior images (SCBI), students’’ basic information, and academic background information to provide a more comprehensive prediction of student academic performance. This study involves training three deep learning models to extract student behavioral features from SCBI, which are classified into positive and negative categories. The behavioral features are then quantified as student engagement, which is then fused with educational information to form the Behavior_Education_FBE dataset. Finally, the Behavior_Education_FBE dataset was used to train five machine learning classifiers to predict whether students can pass the final exam or not. In this study, the Accuracy ( \(ACC\) $\mathit{ACC}$ ), Recall ( \(Rec\) $\mathit{Rec}$ ), and F1 score ( \(F1\) $F1$ ) parameters are utilized to assess the classification performance. According to the experimental results, BISAP significantly outperforms the model that utilizes educational information alone for student academic performance prediction. Among them, Cascade R-CNN and LightGBM models perform best among deep learning models and machine learning models, respectively, and the combination of the two achieves the best prediction performance ( \(ACC = 0.9394\) $ACC=0.9394$ ,  \(Rec = 0.8667\) $Rec=0.8667$ ,  \(F1 = 0.9286\) $F1=0.9286$ ).