Abstract

Facial emotion recognition is a rapidly growing field in artificial intelligence, with applications in human-computer interaction, security, and psychological analysis. The effectiveness of such systems relies on deep learning models that accurately classify emotions based on facial expressions. This study evaluates different deep learning architectures, with Convolutional Neural Networks (CNNs) emerging as the most effective approach due to their ability to capture spatial hierarchies and extract meaningful features. Additionally, the impact of various activation functions on classification accuracy is explored to optimize the model’s efficiency. Experimental results confirm that activation function selection significantly influences CNN performance, affecting both learning stability and classification accuracy.