Autism Spectrum Disorder Diagnosis Through Natural Language Processing and ConvCapsule Fusion Network

The disease termed as autism spectrum disorder (ASD) distracts a person's physical, mental, social, and emotional health. Communication difficulties, social interaction difficulties, obsession, and repeated behaviours are some of the symptoms. Early detection of ASD is essential for reducing the disorder's effects through a variety of behaviour, educational, and family-focused therapy. Tradition methods face issues with human subjectivity, scalability and dependency on expert interpretation. In this paper, an advanced natural language processing (NLP)-assisted deep learning (DL) model (ANLP-DL) model for ASD detection model is proposed to address the existing issues. Initially, a dataset of 47,500 entries sourced from medical records, doctor notes, clinical assessments, test outcomes and online sources are gathered. Feature extraction methods like N-gram level vectorizer, GloVe and Term frequency-based enhanced Inverse class frequency (TF-EICF) are used to extract sufficient features. Next, ASD detection process is accomplished using ConvCapsule Fusion Network (ConvCFN) model. The entire implementation is carried out using the Python platform and assessed using diverse metrics. The finding shows that the ANLP-DL model attains maximum accuracy of 98.3%. Accordingly, the ANLP-DL model proves the potential of incorporating DL and NLP in healthcare technology.