Engagement assessment plays a crucial role in understanding and improving the educational experiences of students with autism spectrum disorder (ASD). Traditional methods of engagement assessment heavily rely on subjective observations and manual coding, which can be time-consuming, prone to bias, and lack of precision. In this study, we propose a deep learning-based approach, specifically convolutional neural networks (CNNs) to automatically analyze multimodal data streams, including facial expressions and body movements collected during classroom activities. The multimodal data is preprocessed and fed into CNN for feature extraction, followed by CNN-2 for temporal modeling and engagement prediction. The results demonstrate that the proposed deep learning-based approach achieves significantly higher accuracy that is 97.1%. In the future, an Intelligent Tutor System (ITS) will be built to help children with ASD study more effectively via the use of mobile or online applications. Evaluating participation from since engagement is a cognitive trait shared by humans, visual signals alone are insufficient. Physiological elements must thus be present. In the future, we will work on a model that can evaluate a student’s degree of involvement by combining behavioral and physiological data.