Real Time Student Emotion Detection using Yolov5

Abstract

The introduction of technology in the field of Education, especially in learner emotion detection plays an important role in the modern educational context. This research introduces the application of the YOLOV5 algorithm to detect learner emotions in real time during the classroom learning process. This research aims to see the performance of YOLOv5 in detecting student emotions by comparing YOLOv5 variants, namely YOLOv5m, YOLOv5n, YOLOv5l, YOLov5s, and YOLOv5x. The dataset used is a video recording of the learning process taken in classroom A3-02 in Building A, Informatics Engineering Study Program, Faculty, Engineering, University of Mataram, which is grouped into 3 classes, namely (Bored, Happy, and Neutral) with a total dataset of 451 images with dataset distribution divided into 87% training data, 8% validation data, and 4% testing data. Based on the tests conducted, YOLOv5m showed the best results with the highest accuracy reaching 89.60% on Mean Average Precision, with batch settings of 14 and epochs of 150. These results indicate that the YOLOv5 algorithm is effective in detecting learner emotions with a satisfactory level of performance and makes a significant contribution to learner emotion detection, underscoring the potential of this technology in enhancing interaction and learning in educational environments.