Transfer Learning Based Autism Detection in Children

Autism Spectrum Disorder (ASD), is a multifaceted neurodevelopmental condition, exhibiting a diverse array of symptoms and varying degrees of severity, which distinguishes it as a particularly complex and unique disorder. Autism affects communication, social interaction, and behavior, often resulting in difficulties related to making eye contact, comprehending social cues, and verbal communication. Initial indicators, like speech and language developmental delays, challenges with maintaining eye contact, and repetitive behaviors, provide valuable diagnostic information. The diagnostic process is aided by the utilization of Facial Image Data. The whole purpose of this work is to enhance the precision of autism diagnosis by using Transfer Learning technique on facial feature classification. To achieve this goal, Famous pre-trained models like ResNet-18 and EfficientNet-B0 were used on Autism Dataset which contains facial images of Autistic and Non-Autistic children. The performance possessed by these models were considerably good, Resenet-18 has achieved highest accuracy of 84.67% on test data and EfficientNet-B0 showing an accuracy of 80.67%.