Feature Signature Discovery for Autism Detection: An Automated Machine Learning Based Feature Ranking Framework

Area of Science:

• Neuroscience and Developmental Psychology
• Artificial Intelligence and Machine Learning

Background:

• Autism spectrum disorder (ASD) is a complex neurodevelopmental condition with varied presentations, impacting social interaction, communication, and behavior.
• Early detection of ASD is crucial for enhancing adaptive skills and independence, yet its exact causes remain elusive and diagnostic methods are evolving.
• Current research indicates a genetic component to ASD, but definitive biomarkers and widely applicable early detection tools are still under development.

Purpose of the Study:

• To apply automated machine learning (AutoML) with feature ranking for identifying significant nonclinical markers for early autism spectrum disorder detection.
• To develop and evaluate a novel AutoML framework that optimizes hyperparameters for ranking potential nonclinical autism markers.
• To report feature signatures and their importance in distinguishing between individuals with and without autism across diverse age groups.

Main Methods:

• Utilized publicly available datasets comprising Q-chat scores from toddlers, children, adolescents, and adults.
• Implemented an automated machine learning framework with hyperparameter optimization for feature ranking.
• Ranked AutoML models using Mathew's correlation coefficient (MCC) and balanced accuracy to identify significant nonclinical markers.

Main Results:

• The proposed AutoML framework achieved approximately 90% MCC and 95% balanced accuracy across all tested age groups.
• Identified and reported significant feature signatures for autism detection, a novel contribution to the field.
• Demonstrated superior performance and flexibility compared to deep learning approaches in marker extraction and adaptability.

Conclusions:

• Automated machine learning offers a flexible and effective approach for early autism spectrum disorder detection by identifying key nonclinical markers.
• The developed framework provides high accuracy and interpretability, facilitating early diagnosis and intervention planning.
• This research highlights the potential of AutoML in advancing diagnostic tools for neurodevelopmental conditions with minimal user intervention.