EMOTION RECOGNITION MODEL DEVELOPMENT USING AUGMENTED-MASKED FERPLUS DATASET

• Main Author: Verel Siedharta, Vincentius
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/82201
• Institution: Institut Teknologi Bandung

Online learning has become one of the teaching alternatives that has received considerable attention since the Covid-19 pandemic. In this research, the author explores solutions to an important problem in online education, namely detecting learner emotions and presenting reports to teachers through a dashboard. The purpose of this research is to develop a model that can detect student emotions through facial expressions despite the presence of obstructing objects. The dataset used in this research is Augmented-Masked FERPlus. The Masked-Augmented FERPlus dataset attempts to solve the problem of dataset imbalance in emotion detection, especially when the face is obstructed by objects such as masks. This dataset is used to train various machine learning models, including Convolutional Neural Networks (CNN), VGG16, ResNet, and InceptionV3. The augmented and masked dataset achieved an accuracy of 0.747 with the Convolutional Neural Networks (CNN) model. Although the augmentation technique used in this study is simple, the experimental results show that this Augmented-Masked FERPlus dataset effectively reduces model overfitting during training.