Facial Emotion Classification Utilizing Efficientnet: A Comprehensive Deep Learning Method

Human-computer interaction has now advanced to rely significantly on emotion recognition through deep learning. This paper studies the capabilities and effectiveness of the EfficientNetV2B2 model for real-time facial expression-based emotion detection in real-world applications. For fine-tuning after training on the FER dataset, I have used the Categorical Cross-Entropy loss function and Adam optimizer. Various data augmentation techniques have been implemented to improve the generalization ability of the model. From our results, it is evident that all seven different emotions, which include anger, disgust, fear, happiness, neutrality, sadness, and surprise, can be distinguished with a great accuracy. Moreover, the architecture of the model ensures rapid inference with suitability for real-time applications like online education, mental health tracking, and so on in the customer support system. At 88%, the deployment results are validated.