Hybrid Particle Swarm Optimized SVM for Neuroimaging-Based Early Autism Identification

Autism Spectrum Disorder (ASD) is a complex neurodevelopmental condition characterized by social, behavioral, and communication impairments. Early detection is critical for effective intervention, yet current diagnostic practices rely heavily on behavioral observations, which are often subjective and time-consuming. This research introduces a Hybrid Particle Swarm Optimization–Support Vector Machine (HPSO–SVM) framework for early and accurate ASD detection using neuroimaging and behavioral datasets. The HPSO algorithm optimizes the hyperparameters of the SVM classifier and simultaneously selects the most discriminative features, improving classification performance. Experimental evaluation was performed on publicly available fMRI and behavioral datasets, including the ABIDE dataset. Quantitative results demonstrate that HPSO–SVM achieved an accuracy of 97.3%, precision of 95.8%, recall of 96.9%, and F1-score of 96.3%, outperforming conventional SVM, PSO-SVM, and Random Forest models. The hybrid optimization reduced feature dimensionality by 42%, improving computational efficiency while preserving discriminative power. Statistical significance tests confirmed the robustness of the proposed approach (p < 0.01). The findings indicate that the integration of bio-inspired optimization and kernel-based learning can effectively capture subtle neuro-patterns associated with ASD, paving the way for automated and objective diagnostic systems.