Optimization and Comparison of Deep Learning Models for Early Detection of Autism Spectrum Disorders Using the TASD Dataset

Abstract

Autism Spectrum Disorder (ASD) is a complex neurodevelopmental condition characterized by challenges in social communication, restricted interests, and repetitive behaviors. Early detection of ASD is critical, as it significantly enhances the effectiveness of intervention and long-term developmental outcomes. However, traditional diagnostic methods—based on clinical observation, parental interviews, and psychological testing—are often subjective, time-consuming, and inaccessible, particularly in under-resourced regions. This research explores the use of Artificial Intelligence (AI), specifically machine learning (ML) and deep learning (DL) models, as innovative solutions to enhance the accuracy, speed, and accessibility of ASD diagnosis. Using the Autism Spectrum Disorder Screening Dataset for Toddlers, the study compares the performance of various ML algorithms such as K-Nearest Neighbors (KNN), Decision Tree (DT), and Random Forest (RF), alongside DL architectures like Artificial Neural Networks (ANN), Recurrent Neural Networks (RNN), and Long Short-Term Memory (LSTM).