Capturing Students’ Attention Through Visible Behavior: A Prediction Utilizing YOLOv3 Approach

One way to determine whether or not the student is conscientious in the classroom is by facial expressions. Facial expressions are facial changes in response to a person's internal mental states, thoughts, or social contact. The application of machine learning and computer vision methods have made very useful in area of automated assessment. In this paper, an experimental setup was installed for data collection. The researchers aim to present a new approach of predicting student behavior (attentive or not attentive) based from face recognition during class session. This demonstrate a real-time detection of student behavior. Using deep learning approach, the acquired data utilized the YOLO (you only look once) v3 algorithm in predicting student behavior inside the classroom. The evaluation was created right after the live feed review. Generated models were tested using mAP to decide which model is appropriate for object detection. The mAP (mean average accuracy) is a common measure used to determine the precision of the artifacts being observed. This measure was focused on the following class: high = Attentive and low = Not Attentive. The experimental testing shows that model accuracy is 88.606%. Tests indicate that this method offers reasonable pace of identification and positive outcomes for the measurement of student interest dependent on observable student actions in classroom instruction.