A Comparative Study on Emotion Analysis Using Transfer Learning

Emotional analysis plays an important role in improving AI's ability to learn and respond to human emotions. Face emotion recognition is expeditiously expanding its field of research with important applications across various domains such as medical image analysis, monitoring, personal identification, and security and theft alerts in current society This technology enhances human-computer interactions, mental health and well-being, market research, education and training, security, and social robotics by enabling machines to recognize, interpret, and respond to human emotions based on facial expressions. Transfer learning has come up with a promising address to leverage pre-trained models for emotion analysis tasks. In this paper, a comprehensive comparative analysis of various transfer learning approaches employed in AI emotion analysis, unveiling valuable insights into their performance. The main objective is to assess the effectiveness and performance of the transfer learning algorithms, namely Resnet-50, Vgg-16, Vgg-19 … etc. To carry on our analysis, we have used the FER2013 emotion dataset. The model's performance is assessed through evaluation metrics which include precision, recall, F1 score, and AUC. Our experimental results demonstrate that transfer learning algorithms significantly enhance the performance of emotion analysis AI systems. Upon stringent experiment and evaluation on FER2013 dataset, our findings reveal that VGG16 demonstrates the highest accuracy.