MULTI-CLASS STUDENT EMOTION CLASSIFICATION IN ONLINE LEARNING USING DEEP LEARNING VIA WEBCAM

Abstract

The traditional classroom learning approach has been replaced with an online learning system in the educational field due to the COVID-19 pandemic. During this pandemic, a lot of students suffered from depression, academic stress, or anxiety in e-learning because they could not adapt to the new norm. Thus, a system that can recognize the students' emotions is beneficial to evaluate the students' emotions during e-learning. Most of the current student emotion recognition systems do not focus on classifying academic emotions. Hence, this project proposed a system that can classify multi-class student emotions based on facial expressions such as "bored", "confused", "engaged", and "frustrated using a deep learning approach. The dataset used in this project is Dataset for Affective States in E-Environments (DAiSEE). Based on the results obtained in the research experiment, it was reported that the CNN model had the highest accuracy (88.92%) compared to LSTM (79.17%) and ConvLSTM (85.33%). This project is expected to develop a real-time system where the student emotions can be evaluated during online learning, and the result of student emotion classification will be shown.