Independent Vector Analysis for Resting-State EEG Classification in Autism Spectrum Disorder

This study investigates Independent Vector Analysis (IVA) for the joint analysis of resting-state electroencephalography (EEG) recordings from individuals with autism spectrum disorder (ASD) and neurotypical controls. The proposed pipeline organizes EEG data into group-wise datasets, estimates IVA-derived latent representations, extracts temporal, spectral, and nonlinear features, selects features using mRMR, and classifies subjects using a Support Vector Machine. In experiments involving 28 subjects, the best IVA-based configuration achieved an accuracy of 75,4\%. The most recurrent selected features were predominantly spectral, suggesting that IVA-derived representations may emphasize information relevant to ASD versus control discrimination.