Improving Student Behavior Prediction and Classroom Interaction Optimization in Hospital Financial Management Course by Deep Transfer Learning Model

There is a problem of data imbalance in the prediction of student behavior in hospital financial management course, and the classroom interaction is highly complex. This paper combines ViT (Vision Transformer) with transfer learning to propose a new student behavior prediction and classroom interaction optimization method to improve the classroom efficiency of hospital financial management course. Using the knowledge obtained from known fields by transfer learning, ViT is used to extract the characteristics of student behavior patterns, and student behavior is predicted. Fine-tuning is performed on the target dataset to optimize the prediction ability of the model and reduce the impact of data imbalance; combined with the results of student behavior prediction, a classroom interaction optimization strategy is designed. By dynamically adjusting the classroom content and teaching methods, the best match with student behavior is achieved, which promotes students’ active participation and improves classroom efficiency. Experimental results show that the cross-entropy loss of the model combining ViT and transfer learning remains low in 10 folding subsets, between 0.20-0.25, showing strong prediction ability, and the participation rate is stable above 0.8, showing a high classroom interaction effect.