With the emergence of new communication and entertainment technologies, the educational system suffers from the ever-increasing number of cheating cases in ^exams. Today, most students have become lazy and want to pass the exam without making a substantial effort in their exam preparation. Consequently, several cheating techniques have emerged, and classical surveillance in the exam has become obsolete. There-fore, the necessity to leverage edge technologies for automating cheating case detection becomes a must. This paper proposes an anti-cheating model focusing on student behavior analysis by analyzing the student's posture in real-time using deep learning and computer vision techniques. We first extract high-level domain features from video frames to realize this goal using CNN-Facial-landmark and media pipe models. Then, we use an LSTM classification model for cheating case prediction. The implementation outcomes show that the proposed model accuracy on the train set reached 94%, and the test set achieved 75%.