A new machine learning model based on induction of rules for autism detection.

Area of Science:

• Developmental Psychology
• Artificial Intelligence in Healthcare
• Machine Learning for Diagnostics

Background:

• Autism spectrum disorder (ASD) presents challenges in communication, social skills, and behavior.
• Current ASD diagnostic methods can be lengthy, costly, and user-dependent.
• Existing screening tools like AQ and M-CHAT have limitations in accuracy and efficiency.

Purpose of the Study:

• To develop an advanced machine learning (ML) classification system for improved ASD detection.
• To introduce Rules-Machine Learning (RML) for automated ASD trait identification.
• To provide interpretable knowledge bases (rules) for understanding classification decisions.

Main Methods:

• Proposed a novel machine learning algorithm: Rules-Machine Learning (RML).
• Developed automated classifiers for detecting autistic traits in diverse age groups.
• Evaluated RML performance against established ML techniques (Boosting, Bagging, decision trees, rule induction) on three datasets.

Main Results:

• RML demonstrated superior predictive accuracy compared to other ML approaches.
• Classifiers generated by RML exhibited higher sensitivity and specificity.
• The method provided a higher harmonic mean, indicating balanced performance.

Conclusions:

• Rules-Machine Learning offers a significant advancement in the accuracy and efficiency of ASD detection.
• The interpretable rules generated by RML can aid domain experts in understanding diagnostic classifications.
• RML shows promise for improving clinical and research applications in autism diagnostics.