Facial Expression Detection System Based on Convolutional Neural Network (CNN) for Human Emotion Recognition

Abstract

The development of human facial expression detection systems has become a growing research topic, particularly in efforts to create applications capable of automatically understanding and responding to human emotions. This research aims to develop and evaluate a human facial expression detection system using the Convolutional Neural Network (CNN) method. The dataset used consists of facial images with various expressions sourced from diverse origins. The data undergoes several preprocessing stages, including normalization, augmentation, and splitting into training and test sets. This study employs several CNN architectures to identify emotions such as happy, sad, angry, and scared. Testing is conducted using various parameters, including training and test data splits, as well as different CNN architectures. The results show that the CNN model can achieve over 90% accuracy on training data, with the best performance on the "Happy" emotion, achieving an f1-score of 0.93. However, there is a decrease in accuracy on validation data, with an overall average accuracy of 78%, indicating challenges in model generalization. Additionally, the "Sad" emotion has the lowest recall of 0.49, indicating the need for model improvement in classifying specific emotions. This study contributes to the development of CNN-based facial expression detection systems, but further exploration of more complex architectures, evaluation with diverse datasets, and real-time testing are needed to improve system performance.