Enhancing Early Autism Diagnosis and Personalized Interventions in the U.S.: AI-Driven Approaches with Implications for Special Education Policy and Practice

Abstract

Autism Spectrum Disorder (ASD) diagnosis faces significant challenges due to reliance on subjective behavioral assessments, leading to delays in early intervention. This research aims to enhance ASD diagnosis and personalized interventions using AI-driven machine learning models. The study utilizes the Autistic Spectrum Disorder Screening Data for Adults, consisting of 704 observations and 21 attributes, including demographic and behavioral features. Focusing on the U.S., Logistic Regression, LightGBM, and Multi-layer Perceptron (MLP) models are employed to optimize predictive accuracy. Logistic Regression achieved the highest balanced accuracy of 88.65%, outperforming LightGBM (87.94%) and MLP (88.00%). T-tests showed significant differences in model performances: Logistic Regression vs. LightGBM (t = 6.96, p = 0.00012) and Logistic Regression vs. MLP (t = 6.07, p = 0.00030). However, Wilcoxon tests revealed no significant differences between models (p = 0.0625). LightGBM exhibited the highest precision (0.91) but suffered from a lower recall (0.77), which is critical for ASD detection. Despite AI's high accuracy in controlled environments, real-world applicability is hindered by dataset homogeneity, with accuracy dropping from 86% to 72%. Additionally, only 33% of AI-based ASD tools integrate with Electronic Health Records (EHR), limiting clinical adoption. These findings highlight the need for AI-assisted, clinician-approved diagnostic tools, enhanced dataset diversity, and real-world validation to ensure ethical and scalable AI deployment in ASD screening. AI-driven personalized interventions, such as wearable-assisted therapies, show promise in improving ASD-related social engagement by 25%, yet clinician skepticism remains high, with 62% expressing concerns about AI reliability without human oversight. Future research should focus on improving model transparency, addressing bias, and integrating AI into existing clinical workflows to maximize its impact in ASD diagnosis and treatment.

Ähnliche Arbeiten