A clustering approach for autistic trait classification.

Area of Science:

• Neuroscience
• Computer Science
• Medical Informatics

Background:

• Autism Spectrum Disorder (ASD) screening relies on accurate detection of characteristic traits.
• Existing screening tools often use scoring functions, which may not capture the full spectrum of ASD presentation.
• Machine learning (ML) offers potential for enhancing the efficiency and accuracy of ASD screening.

Purpose of the Study:

• To develop and validate a novel semi-supervised ML framework, Clustering-based Autistic Trait Classification (CATC), for improved ASD trait detection.
• To reduce data dimensionality and eliminate redundancy within autism datasets to optimize ML model performance.
• To identify potential autism cases based on trait similarity rather than solely relying on scoring mechanisms.

Main Methods:

• Proposed a semi-supervised ML framework named Clustering-based Autistic Trait Classification (CATC).
• Employed clustering techniques to group individuals based on trait similarity.
• Validated classifiers using established classification techniques.
• Reduced data dimensionality and redundancy in autism datasets.

Main Results:

• CATC demonstrated higher predictive accuracy compared to Artificial Neural Network (ANN), Random Forest, Random Trees, and Rule Induction.
• The proposed method achieved superior sensitivity and specificity rates in identifying ASD traits.
• Empirical results were verified across diverse datasets encompassing children, adolescents, and adults.

Conclusions:

• The CATC framework offers a promising approach for enhancing the accuracy and efficiency of Autism Spectrum Disorder screening.
• Identifying potential autism cases through trait similarity provides a valuable alternative to traditional scoring functions.
• The developed ML classifiers are beneficial tools for diagnosticians and stakeholders involved in ASD screening.