Autism Spectrum Disorder Detection In Early Ages Using Automated Machine Learning

Abstract

Autism spectrum disorder (ASD) is a neurodevelopmental condition characterized by a diverse range of symptoms and levels of ability. Detection of ASD at early ages is desirable as it allows for early intervention, which can improve the child's condition. However, the conventional diagnostic process requires hours of clinical examination, which can be time-consuming and expensive. The thesis proposes the use of auto-ML as a tool to simplify the diagnosis process and improve precision. The study collected data from multiple rehab centers in Pakistan and applied the auto-ML framework TPOT to the dataset for ASD detection. The results showed that TPOT gave the best pipeline for the dataset, with the highest accuracy of 79%, and it was verified. The study contributes to the field of ASD diagnosis by utilizing auto-ML to identify the likelihood of ASD in children during the early stages of development. The study also provides an evaluation of precision, recall, and F1-Score metrics to verify the accuracy of the diagnosis. Overall, this thesis presents a promising approach to improve the detection of ASD in children, which can ultimately lead to better outcomes for affected individuals and their families.