Autism Spectrum Disorder Prediction Using Ensemble Learning

Abstract

Autism Spectrum Disorder (ASD) is one of the dominant complex neurodevelopmental situations that presents a global health challenge due to its increasing prevalence and substantial impact on quality of life. Effective intervention depends on an accurate and fast diagnosis, however traditional diagnostic techniques can be expensive and limited by the availability of clinical resources. Compared to traditional methodologies, machine learning (ML) techniques have shown the potential to predict ASD risk with more accuracy and reliability. In this study, ML-based techniques such as Decision Trees (DT), Random Forests (RF), and XGBoost (XG) are investigated on datasets that include behavioral, clinical, and demographic traits. The models outperformed conventional diagnostic methods in terms of accuracy and sensitivity, achieving noteworthy crossvalidation accuracies with decision trees at 79%, random forests at 87% and XGBoost at 90%. To enable customized, real-time risk assessments, the usage of ML with realtime instruments like wearables and Electronic Health Records (EHR) are also investigated. These results demonstrate how ML can increase the faster identification and treatment of ASD, opening up possibilities for more affordable and effective diagnostic options.