Autism spectrum disorder is a neurological illness. Social interaction, communication and behavior are all impacted. For prompt intervention and improved care, early and precise detection is crucial. Traditional diagnostic procedures are based on behavioral assessments, which can be subjective and time-consuming. Automating the identification of ASD from facial visuals has showed encouraging results with Convolutional Neural Networks, a type of deep learning. Individuals with ASD frequently exhibit social and emotional processing problems, which can be indicated by subtle facial movements, eye gazing patterns, and facial emotions. By enhancing model accuracy using sophisticated CNN architectures and optimal preprocessing methods, this study expands on earlier studies. This study contributes to the growing corpus of research on combining AI-powered technologies with conventional diagnostic techniques to improve the speed and precision of ASD identification, which will eventually benefit those on the autistic spectrum.