Autism Detection in Children: Integrating Machine Learning and Natural Language Processing in Narrative Analysis

Area of Science:

• Computational linguistics
• Developmental psychology
• Machine learning

Background:

• Early identification of autism spectrum disorder (ASD) is critical for improved outcomes.
• Diagnosis in Greece averages six years, with further delays for lower-income or minority families.
• Current language assessments for ASD are time-consuming and labor-intensive, hindering early diagnosis.

Purpose of the Study:

• To develop a reliable and practical artificial intelligence (AI) model for early ASD identification.
• To distinguish children with ASD from typically developing peers using narrative and vocabulary skills.
• To create objective tools that increase access to early ASD assessment and diagnosis.

Main Methods:

• Applied natural language processing (NLP) to extract language features from storytelling.
• Utilized machine learning models trained on narrative and vocabulary data from 68 children with ASD and 52 typically developing children.
• Compared performance of decision trees, gradient boosting, hist gradient boosting, and XGBoost models.

Main Results:

• The AI model achieved 96% accuracy in distinguishing children with ASD from typically developing peers.
• Hist gradient boosting and XGBoost models demonstrated superior performance in accuracy and F1 score.
• Language features extracted via NLP effectively differentiated between the two groups.

Conclusions:

• AI-powered language analysis provides a rigorous and objective method for ASD identification.
• This technology can significantly improve early diagnosis accessibility, especially for underserved populations.
• Machine learning deployment holds promise for enhancing early identification services for children with ASD.