Deep Learning Framework for Autism Spectrum Disorder Detection: A CNN-Based Comparative Analysis

People with Autism Spectrum Disorder (ASD) struggle with social contact and communication for the rest of their lives. Because symptoms often emerge in the initial two years of life, autism is sometimes referred to as a “behavioural disease” and may be diagnosed at any point in a person's life. As far as ASD is concerned, the issue does not go away after a kid enters puberty or enters adulthood This study investigates the application of machine learning and deep learning techniques for the early detection of Autism Spectrum Disorder (ASD) across children, adolescents, and adults. Six classification models Naïve Bayes, Support Vector Machine, Logistic Regression, K-Nearest Neighbors, Artificial Neural Network, and Convolutional Neural Network are evaluated using three publicly available ASD screening datasets that do not include any clinically relevant ASD information are used to assess the suggested methods. The first dataset for autism spectrum disorder screening for kids has 292 occurrences and 21 variables. Seven hundred four occurrences with twenty-one variables make up the second dataset pertaining to ASD screening in adults. The third collection of data pertains to screening for autism spectrum disorder in adolescents, and it has 104 cases with 21 characteristics After data pre-processing and model optimization, experimental results demonstrate that CNN consistently outperforms traditional models, achieving accuracies of 99.53% (Adults), 98.30% (Children), and 96.88% (Adolescents). The findings highlight the effectiveness of deep learning models in supporting reliable and automated ASD screening.