Analyzing Limitations of Pre-Trained Deep Learning Models for Facial Emotion Recognition

Kajipuram, Harshavardhan (2024) Analyzing Limitations of Pre-Trained Deep Learning Models for Facial Emotion Recognition. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (820kB) | Preview
Preview	PDF (Configuration Manual) Download (301kB) | Preview

Abstract

Deep learning facial emotion recognition is one area of research that has attracted considerable interest in recent years because of its usability across healthcare, human-computer interaction, and security. This work evaluated the drawbacks of using pre-trained deep learning models, especially the VGG16, towards FER applications. The research hypothesis was to establish the rigidity factors that limit model generalization and examine how performance can be enhanced with more realistic test set restrictions, including class imbalance, diversity, and variations in facial expressions.

The VGG16 model was fine-tuned and extensively trained using the FERC dataset, which consists of seven emotion classes: militant, Disgust, fear, happiness, neutral, sadness, and surprise. The hyperparameters in the models were tuned using the learning rate, dropout rate, batch size, and the type of optimizer used (Adam, SGD, RMSprop). While the training accuracy exceeded 80%, the validation accuracy stagnated at 32.0%, indicating significant overfitting.

The study identified key limitations, including dataset quality, class imbalance, and the complexity of subtle emotional features. Recommendations include exploring alternative architectures such as ResNet, EfficientNet, and Vision Transformers (ViT), enhancing datasets through augmentation and balancing, and incorporating advanced evaluation metrics like precision and F1-score. This research highlights the need for further improvements to address overfitting and ensure robust performance in practical applications.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Nagahamulla, Harshani UNSPECIFIED
Subjects:	Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science Q Science > QH Natural history > QH301 Biology > Methods of research. Technique. Experimental biology > Data processing. Bioinformatics > Artificial intelligence > Computer vision Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Artificial intelligence > Computer vision B Philosophy. Psychology. Religion > Psychology > Emotions
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	02 Sep 2025 15:14
Last Modified:	02 Sep 2025 15:14
URI:	https://norma.ncirl.ie/id/eprint/8720

Actions (login required)

View Item