Transfer Learning and Hybrid Deep Convolutional Neural Networks Models for Autism Spectrum Disorder Classification From EEG Signals

Autism spectrum disorder (ASD) is a developmental disease characterised by restricted and repetitive behaviours, as well as difficulty in social communication and interaction, in children. The clinical diagnosis of ASD is reached by behavioural screening, which delays early intervention. Electroencephalography (EEG) is a method for analysing the brain's electrical activity that has proven useful in the diagnosis of several neurological illnesses. Pre-trained deep Convolutional Neural Networks (CNNs) were used to extract features from the spectral profiles of the EEG dataset and classify patients into mild, moderate, and severe patients, as well as age-matched control subjects. Accordingly, the primary goal of this study is to use the pre-trained CNNs as classifiers in order to reap the benefits of transfer learning, and the secondary goal is to propose a hybrid model by employing decision tree (DT), K nearest neighbour (KNN), and a Support Vector Machine (SVM) machine learning classification techniques to categorise the features of the pre-trained CNN networks into mild, moderate, severe, and normal categories. The results show that using SqueezeNet for transfer learning improves classification accuracy to 85.5%, and that using SqueezeNet for hybrid models improves classification accuracy to 87.8% using SVM. Therefore, a hybrid model based on the combination of SqueezeNet and SVM might be utilised to automatically diagnose ASD based on the individual's EEG data.