Leveraging Transfer Learning with Stacked Ensemble Learning for theDetection of Autism Spectrum Disorder using Eye-tracking

A neurological condition known as Autism Spectrum Disorder (ASD) has an impact on various facets of children’s behavior. Some Children show symptoms within 12 months, forsomeothers it may take even after 2 or 3 years. Some people can go undetected until they reach adolescence.Whenautisticindividuals grow into adults, they may often have more problems interacting with peers, difficulty in comprehending the actions of others, or related to attention. This causes anxiety, depression, or other problems, making life difficult. Early identification and intervention are vitally significant. Recent developments in Eye tracking technology have given medical professionals useful tools to improve the precision and timeliness of diagnosis. Deep neural networks (DNN) have shown outstanding performance in ASD identification and numerous image-related tasks, their architectures confront many difficulties, including reduced accuracy, increased complexity, and difficulties in extracting strong features. This study utilized a pretrained Xception Transfer Learning (TL) model for feature extraction followed by dimensionality reduction. A two-stage stacking ensemble framework is developed for autism classification. The proposed Stacked Ensemble model demonstrated the great potential for correctly identifying Autism with an accuracy of 98.9% and an AUC measure of 99.0%.