Implementation of Face Emotion Detection In Classroom Using Convolutional Neural Network

Emotion detection is one of the most complex and challenging computer visuals due to significant variations caused by changes in facial appearance, lighting, and facial expressions. The results of tests conducted by Amanullah et al. on the classification of emotions using the Convolutional Neural Network show an accuracy value of 81.92% for training and 81.69% for testing but still has a fairly large error rate of 75.60% based on the loss validation results. In this study, the research to be conducted is to study more deeply about detecting human emotions based on facial expressions using the Convolutional Neural Network method. To detect human emotions based on faces using a Canon EOS 60D camera with good output quality. The camera is placed in front of the class. The computer user uses the Windows 10 64bit operating system, with a Core i5 processor, 16BG RAM, and GTX1060 VGA. Four LED lights are used every 18 watts to overcome the room lighting. Every 60 seconds, the system will carry out the process of detecting and classifying emotions and storing them in the database. The data obtained from the test results will be analyzed to obtain data in the form of accuracy, precision, and recall. The dataset for the training data used is FER-2013 which contains 28,000 datasets taken from Kaggle. The research results obtained were to obtain a high accuracy value that exceeded the results tested by Amanullah et al. The research results that will be carried out are also expected to be used by teachers to detect students' emotional conditions during class learning.