AI-Powered Autism Screening: Leveraging Gain Ratio Feature Selection for Improved Machine Learning Models

Early screening of Autism Spectrum Disorder (ASD) in toddlers is critical for timely intervention, yet traditional diagnostic methods often lead to delays. This study introduces an integrated machine learning (ML) framework to enhance ASD screening in toddlers aged 12 to 36 months. The proposed framework incorporates Gaussian Noise Up-Sampling to address class imbalance, the Normalizer for feature scaling, and the Gain Ratio Attribute Evaluator (GRAE) for feature selection, ensuring optimal input for classification models. Six ML models—Adaptive Boosting, Extreme Gradient Boosting, Random Forest, K-Nearest Neighbors, Naïve Bayes, and Logistic Regression—were applied to a merged dataset comprising diverse cultural backgrounds. Results demonstrate outstanding classification performance, with Extreme Gradient Boosting achieving 100% accuracy, precision, recall, F1, and kappa when using GRAE-selected features. The findings emphasize the importance of feature engineering in refining ASD-related behavioral attributes while eliminating redundancies. The proposed framework presents a scalable and culturally adaptive tool for ASD screening, addressing diagnostic delays and enabling timely interventions. By integrating ML into early detection strategies, this approach has the potential to improve developmental outcomes for toddlers worldwide.

Перевод пока недоступен