Optimized Pre-trained Feature Selection with Support Vector Machine Model For Autism Spectrum Disorder Detection

Abstract

Autism Spectrum Disorder (ASD) is associated with a nervous system development condition identified by persistent challenges in social communication and behavior, typically identified in the formative years. It is associated with repetitive behaviours and difficulties in social interaction among affected beings. Various approaches to autism spectrum disorder classification have been developed, comprising emotional tests, facial image analysis, and neuroimaging techniques. ASD is a challenging task to diagnose through medical analysis, and some tests are time-consuming and more expensive. In this research article, a novel approach with an Optimized Pre-Trained Feature Selection With Support Vector Machine (OPFSVM) detection model for ASD is proposed to overcome the existing challenges and problems. The novel method accurately identifies Autism in children; this study employed a pre-trained ResNet50 feature extraction method. The feature selection process is performed using a Particle Swarm Optimization (PSO)approach that helps improve system performance by eliminating irrelevant feature sets while retaining the most significant ones. Subsequently, the Support Vector Machine (SVM) model is applied to perform two-class classification of ASD. The proposed (OPFSVM) model integrates pretrained feature extraction and optimized feature processing in the SVM model with binary classification to accurately detect Autism in children. For training the proposed model, an online accessible dataset is used, including facial images for kids, analyzed with Autism, and control subjects categorized as either autistic or non-autistic. According to the outcomes, the suggested OPFSVM model is achieving 97% accuracy, 97% precision, and reducing the 3% error rate, compared with other methods (Vgg19, ResNet50, MobileNet, etc.). These findings highlight the implemented method's high effectiveness in early ASD detection and position it as an effective tool for timely and rapid analysis.