Autism Spectrum Disorder Classification using Multi Scale ResNet based on Facial Expressions

Autism Spectrum Disorder (ASD) is a neuro developmental illness that affects behavior, speech, and social interaction. An early and correct diagnosis is essential for prompt action and assistance. The XGBoost classifier and a Multi-Scale ResNet architecture are used to provide a unique framework for ASD classification focusing on facial expression analysis. By efficiently capturing characteristics at several sizes, the Multi-Scale ResNet catches subtle variations in facial expressions symptomatic of ASD. The XGBoost classifier, known for its performance and scalability, processes these features to achieve classification with an accuracy of 95%. The approach leverages XGBoost's classification abilities and ResNet's feature extraction to address problems like feature redundancy and classification imbalance. Experimental evaluation on benchmark datasets shows the framework outperforms existing methods in accuracy, sensitivity, and specificity. The proposed technology offers a feasible, non-invasive alternative for ASD screening in clinical and educational settings.