In traditional learning situations, teachers mainly evaluate students’ behavioral changes, emotional changes, and interactions to ensure teaching quality. The Random Forest algorithm is employed in this paper to determine the characteristics of students’ body postures and observe their behavioral changes. The geometric analysis method is used to recognize students’ head posture, and the image test classification module is used to capture students’ facial expressions. Analyze the teacher-student learning interactions in the classroom by examining the student’s question-answering records from the interactive cloud platform. A multimodal information fusion model was constructed to combine students’ multimodal performance data to complete the evaluation of the preschool hygiene course. The results of the case study of five students showed that the fusion results of the ratings of students 1-5 under the multimodal information fusion model of the course evaluation were 0.56, 0.888, 0.083, 0.452, and 0.957, respectively. The results of the fusion of the ratings of students 2 and 5 were very serious in learning the course, and the percentage of their time spent in attentive behavior was 0.452. Student 2 and Student 5 studied the course very seriously, spent 90% and 89% of their time engaged in attentive behaviors, respectively, and maintained an emotion of interest for 66% and 70% of the time, respectively. Furthermore, Student 2 answered all the questions in the interactive session, while Student 5 answered all the questions correctly. The median integration rating of the 40 participating students was 0.706, and the majority of the students rated the integration of the course in the middle to high range, which means that the content and format of this pre-school hygiene course were excellent and stimulating for the students.