Machine Learning-Based Autism Spectrum Disorder Detection Using EEG Signals

Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder that impacts the communication, social interaction, and behavior of children. It is critical to diagnose at an early stage. Conventional diagnostic procedures are available based on subjective assessments. This paper aims to design a machine learning (ML) model to diagnose ASD. The input to the model is the Electro Encephalogram Signal (EEG) and the model output classifies whether the children are affected with ASD. Support Vector Machine (SVM), ML model implemented to this binary classification. To train the model 600 EEG signals and each of 300 seconds were collected. In the feature extraction stage, each EEG signal is decomposed with several frequency bands Delta wave (0.5-4HZ), Theta wave (4-8Hz), Alpha wave (8-12Hz), Beta wave (12-30Hz), Gamma wave (30-70Hz) using digital filter bank. The data set is labeled, by analyzing the behavior using Shannon Entropy, Sample Entropy and Lyaponov exponent. Feature selection is carried out using Mutual Information (MI) to find the most appropriate biomarkers. The SVM in trained and validated using K-Cross folding method. Implementation results shows that the accuracy of the model is 92.5 percentage. Experimental results prove high classification accuracy, validating the efficacy of the proposed method. The current research demonstrates the promise of machine learning models based on EEG as a scalable and objective measure for early diagnosis of ASD, enhancing clinical decision making and treatment plans.