Comparative studies of facial emotion detection in online learning

Due to the COVID-19 outbreak that hits everyone globally, every person is affected, including students. Starting with this, no one can refuse the importance of a smart online learning system being embedded in the education system anymore. Still, the emotional engagement between teachers and students...

Full description

Saved in:
Bibliographic Details
Main Authors: Ahmad, Asraful Syifaa’, Hassan, Rohayanti, Zakaria, Noor Hidayah, Moi, Sim Hiew
Format: Conference Proceeding
Language:English
Subjects:
Algorithms
Classification
Comparative studies
Datasets
Distance learning
Embedding
Emotion recognition
Emotions
Image resolution
Machine learning
Students
Teachers
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Due to the COVID-19 outbreak that hits everyone globally, every person is affected, including students. Starting with this, no one can refuse the importance of a smart online learning system being embedded in the education system anymore. Still, the emotional engagement between teachers and students is in doubt. Teachers tend to teach without knowing whether the students understand it. It is due to the limitation of having an online class instead of a physical one, but can be overcome by embedding emotion detection using facial expressions during the course. However, the next challenge is to overcome the system’s limitations. Even though people have a well-equipped tool for themselves in this situation, it is hard to determine the exact emotion of the students during the learning process. The facial image captured in the real world usually has multiple issues, including (1) an obstacle in front of the face, (2) a different distance between the head and camera, and (3) low-resolution images due to the low bandwidth, causing the low accuracy of emotion classification. Thus, this paper took the initiative to compare two different datasets tested on four classifiers: CNN, DCNN, Transfer Learning, and Multiple Pipeline. Four different algorithm sets were used to test two datasets: FER2013 and our dataset. The result shows that the data captured in a real-world situation using an independent device setting contains some issues with the slightly low accuracy of emotion classification and leads to false classification compared to FER2013 data. Thus, this can be improved by having a powerful emotion detection system that can capture all the issues in a real-world situation
ISSN:0094-243X
1551-7616
DOI:10.1063/5.0164746