Exploring Emotion Recognition in Children with Autism Spectrum Disorder using ChatGPT

This study aims to develop and evaluate a neural network-based intelligent diagnostic model for identifying different emotional states in children with autism, including calmness, happiness, sadness, and anger. We utilized a dataset containing images of children with autism's expressions, which was divided into training, validation, and testing sets. Data preprocessing included Gaussian modeling and data encoding to enhance the model's efficiency and accuracy. By leveraging ChatGPT for preprocessing of voice signals and generating spectrograms, combined with facial expression analysis, we gained a deeper understanding of the emotional states of children with autism. The study's results indicated that the model performed well in recognizing calm and happy emotions but underperformed in identifying sadness and anger. Model performance was assessed using accuracy, precision, recall, and F1 score, with an overall accuracy of 0.70, precision of 0.39, recall of 0.39, and F1 score of 0.40. The findings of this study provide a new perspective in the field of emotional recognition for children with autism and lay the groundwork for future research and model optimization.