Early Screening of Autism among Children Using Ensemble Classification Method

Autism Spectrum Disorder (ASD) is a neurodevelopmental disability that affects the nonverbal and verbal communication and social interaction significantly noticeable before the age of three that skeptically affects a child's overall development. Some prominent autistic features include repetitive verbal and movement patterns, as well as atypical responses to neurological stimuli. [2]. The prevalence of disorders at early childhood will help to support the individual with required interventions or therapy. Medical professionals can diagnose developmental disorders when the parents approach the clinicians. Usually screening tools can help parents or caretakers to identify these symptoms at an early stage. Use of Deep Learning techniques proved to be potential in screening of autistic traits through interaction with electronic devices used by the kids at an early stage. The model proposed is based on both Image and Video processing. Initially, A CNN is built for classifying the facial images which pre-processes the images and detects facial landmarks from the given images and trained with the Autistic Children Dataset. For Emotional Responsiveness the CNN is trained with FER2013 dataset and is analyzed by playing a video and response of child is captured by an inbuilt webcam and is processed to extract the facial expression. Also, focus of the child is anticipated by using an eye tracking tool which captures frames, wherein the child is asked to watch the video being played. The performance of the model designed is measured using Machine Learning metrics like Sensitivity, Specificity, AUC-ROC Curves. The accuracy is around 84% and the model is capable of classifying emotions Happy, Angry, Sad, Neutral, surprise better compared to Disgust and Fear. Simultaneously Eye Gaze is also tracked to know the focus of the child and is predicted to be around 65%.