Improving Emotional Well-Being in Autistic Children through Therapy Selection

Abstract

Introduction: Autistic children frequently struggle with recognizing and expressing emotions, highlighting the need for customized therapeutic approaches based on their specific emotional states. This research aims to map appropriate therapies by analyzing emotions detected through EEG signals and facial expressions, shifting the focus from autism or emotion detection alone to therapeutic recommendations. The study specifically targets emotions like happiness, sadness, anger, and anxiety to determine the most effective interventions.

Objectives: The study has three primary goals: first, to classify therapeutic interventions into behavioral, socio-emotional, and neurological categories for precise emotion-based recommendations; second, to develop an ASD detection framework using EEG and facial data to identify emotional states; and third, to compare the effectiveness of EEG signals and facial image analysis in guiding therapy decisions.

Methods: The research employs EEG signals and facial image data processed through machine learning models, including LSTM, CNN, DT, SVM, KNN, and RF. Therapeutic interventions are grouped into behavioral (e.g., ABA, CBT), socio-emotional (e.g., music therapy, sensory integration), and neurological (e.g., neurofeedback, brain stimulation) approaches. A Flask-based web application integrates these models, enabling users to input EEG or facial data for real-time emotion prediction and therapy suggestions.

Results: EEG-based classification achieved exceptional accuracy, with CNN, DT, KNN, and RF models reaching 100% accuracy. Facial image analysis, however, showed more modest results, with CNN achieving the highest accuracy of 47.3%. The therapy mapping system successfully linked detected emotions to tailored interventions, such as neurofeedback for anxiety. The Flask application demonstrated practical utility by providing a seamless interface for input and recommendations.

 Conclusions: The findings underscore the superiority of EEG signals over facial analysis for emotion classification and therapy mapping in autistic children. The proposed framework effectively combines diagnostic and therapeutic tools, offering a practical solution for personalized care. Future research could explore multimodal approaches, integrating EEG and facial data, to further enhance the system's accuracy and robustness.