Abstract :

Autism Spectrum Disorder (ASD) Is A Neurodevelopmental Disorder In Which Early Identification And Proper Assessment Can Make A Great Difference. The Use Of Machine Learning Techniques For Classifying And Detecting ASD Has Become One Of The Hotspots In The Medical Field. However, It Is Still Challenging To Develop Highly Performing But Also Interpretable Predictive Models. This Study Proposes An Explainable Hybrid Machine Learning Framework For ASD Screening Using Behavioral And Demographic Information. The Study Utilizes A Publicly Available ASD Screening Dataset Containing 704 Records And 21 Original Attributes. After Duplicate Removal, Missing-value Handling, Feature Processing, And Data Quality Assessment, 699 Records Were Retained For Modeling. Logistic Regression, Support Vector Machine, Random Forest, And XGBoost Were Evaluated, Followed By Hyperparameter Optimization Of XGBoost Using Optuna. An Ablation Study Was Conducted To Investigate The Contribution Of Behavioral And Demographic Feature Groups. The XGBoost Model Demonstrated Exceptional Performance, Achieving An Accuracy Of 97.86%, Precision Of 94.74%, Recall Of 97.30%, And An Impressive F1-score Of 96%. The Area Under The ROC Curve (ROC-AUC) Came Up To 99.84%, While The Area Under The Precision-recall Curve (PR-AUC) Reached 99.59%. Besides, The Models Brier Score Was Only 0.0167, Which Indicates Outstanding Calibration. The Most Informative Behavioral Features According To The SHAP Analysis Are A9, A6, A5, A7, And A4. Local SHAP Explanations And Counterfactual Analysis Were Utilized To Generate Instance-specific Interpretations Of The Models Predictions. The Results Indicate That Behavioral Screening Responses Offer Significantly Better Predictive Insights Than Demographic Characteristics In This Dataset. The Suggested Framework Can Thus Function As A Clear Computational Method For ASD Screening, While Clinical Evaluation Continues To Be Crucial For Diagnosis.

Published:

11-9-2026

Issue:

Vol. 26 No. 9 (2026)

Page Nos:

218 - 233

Section:

Articles

License:

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

How to Cite