One of the burgeoning directions in computer vision is a student behavior detector, which has significant applications with regard to classroom interaction, attention, and engagement. It makes improvements in educational outcomes due to instant insights that could inform instructors in changing the delivery approach. Since deep learning has been an emerging trend over the recent years, the accuracy and sharpness of behavior detection have seen improvements over time. The detection systems have improved vastly. Modern techniques of deep learning have allowed such systems to capture subtle cues and patterns in behavior, thereby making them applicable for the automated class monitor, remote learning, and personal feed-back systems. This work is a survey of the literature that deals with the detection of student behavior using models such as CNN, LSTM, YOLO models with specific mechanisms, Dlib, OpenPose, and Enhanced GAN models trained on diverse public datasets. We have assessed the performance of surveyed papers based on evaluation metrics that can measure performance with much efficiency. The core purpose of this survey is to cover all understanding deep learning approaches toward their pros and cons. Finally, at the end, there are future scopes presented for the chances of improvement.