Development of AI Based Autism Detection System

Autism Spectrum Disorder (ASD) is a complex neurodevelopmental condition that impacts individuals of all ages. This disorder comprises of symptoms ranging from social and communicational challenges to an exhibition of repetitive behaviors. Early diagnosis and prognosis of ASD is focal for managing its symptoms, supplementing learning outcomes, and social skill development. Conventional diagnosis of ASD based on behavioral evaluations confronts problems owing to the lack of clear-cut rules. On the other hand, Machine Learning and Deep Learning models-based techniques excel in diagnosing ASD by assessing behavioral factors, finding even the tiniest contrariety in gaze and facial attributes. This paper aims to propose an AI-based Autism Detection System leveraging two distinct models. For children aged 2 to 8, XceptionNet model is employed to assess facial dysmorphology in Autistic children's images with an accuracy of 85%, while individuals aged 9 and above undergo evaluation by Light Gradient Boosting Machine Classifier which considers diverse indicators, such as the ASD score, age, gender, ethnicity, etc. to detect Autism with 99% accuracy. Thus, the proposed models enhance the accuracy of ASD detection by addressing the distinct characteristics and requirements of people of all age groups with Autism.