Ensemble Learning Approach for Early Prediction of Autism Spectrum Disorder in Children

Abstract

This study presents the development and evaluation of an ensemble learning model for the early prediction of autism spectrum disorder (ASD). By integrating machine learning algorithms, the study aims to improve the detection of ASD-positive cases. Using a dataset comprising behavioral and demographic attributes, comprehensive preprocessing, descriptive analysis, and model evaluations were conducted on random forest (RF) and extreme gradient boosting (XGBoost). The results yield a high accuracy of 82% - 84% in identifying non-ASD cases, with F1-scores nearing 90%. However, the model showed moderate sensitivity in detecting ASD-positive cases due to overlapping features and data imbalance. Despite this limitation, the ensemble model outperformed individual classifiers, signifying its effectiveness for real-world screening applications. The study highlights the importance of ensemble learning, in building robust diagnostic systems. It contributes a practical, reproducible framework for integrating machine learning into neurodevelopmental disorder screening, advocating for its adoption in health informatics.