Development of an Electrode Reduction Model to Support ASD Diagnosis Based on EEG and Machine Learning

Autism Spectrum Disorder (ASD) is characterized by deficits in communication and social interaction. Based on this, the need for differential diagnoses is increasingly discussed. This study explores the efficacy of reducing the number of electrodes in EEG for improved diagnosis of Autism Spectrum Disorder using machine learning techniques. Using the Sheffield public database, which contains EEG signals from individuals with ASD, our methodology integrates an innovative interpolation method to estimate missing data and employs an evolutionary search algorithm to select the most pertinent features, thus reducing the initial electrode count of 64 to 17. Among the various trained models, our best classifier achieved accuracy exceeding 93%, reinforcing the hypothesis of discriminative EEG patterns associated with ASD. By enhancing the efficiency and accessibility of ASD diagnosis, this research contributes to early intervention efforts, ultimately improving outcomes and quality of life for affected individuals while streamlining healthcare processes.