A Robust Computational Framework for Autism Spectrum Disorder Identification Using Optimized Image Processing and Hybrid Learning Models

The classification of autism spectrum disorder (ASD) has reached a new stage of development that includes the former machine learning (ML) designs and image analysis designs. The study introduces a new framework that uses discrete wavelet transformation with Gaussian filter kernel (DWT‐GFK) to achieve the robust preprocessing of the image, including noise removal and preserving edges without affecting the quality of the image. The VGG16 deep learning model is used to extract features, elaborate visual patterns, and data features that are pertinent to the identification of ASD. The classification stage uses a family of Elevated Learning‐based Boosting Network with Hybrid Learning with Neural Classifier Logic (ELBN–HLNCL) that is optimized with the help of Walrus Optimization Algorithm (WaOA) to derive optimal model settings. Experimental testing of the proposed methods using the autism image dataset (AID), ASD screening dataset, and ABIDE dataset has shown the superiority of the proposed methodology with 99.95% accuracy, 99.9% recall, 99.8% precision, and 99.85% F 1‐score. The results demonstrate strong competitive performance compared with existing ASD classification approaches across multiple datasets. The findings offer promising applications in early ASD detection, facilitating improved diagnostic tools and aiding healthcare professionals in accurate assessments.