Autism Spectrum Disorder Detection: Subjective and Objective Approaches

Autism Spectrum Disorder (ASD) indicates a condition of human brain development that is related to the person’s social interactions, communication, and genetic disorder. Based on behavioral and neurophysiological data, Subjective and objective classification of machine learning (ML) algorithms have demonstrated encouraging results in assisting in the identification of ASD. Subjective classification is based on the AQ10 dataset, which is a brief self-report questionnaire used to assess autism-related traits in individuals. We collected AQ10 datasets for different age groups from the UCI machine learning repository and applied different ML algorithms such as logistic regression (LR), KNN, SVM, decision tree (DT), and random forest (RF) classifier. Based on these trained models and the AQ10 questionnaire, we built a web application for predicting ASD in people of all ages. After that, we worked with objective classification, where ML classification is done with EEG data. We collected EEG datasets for adult people from Sheffield’s EEG datasets. Logistic regression, decision tree, and random forest classifier were used to build prediction models with the EEG dataset. K-fold cross-validation and grid search techniques were used to optimize the performance of the prediction models. The evaluation result showed that the random forest classifier with k-fold cross-validation provided better accuracy than other proposed models.