A Feasibility Approach in Diagnosing ASD with PIE via Machine Learning Classification Approach using BCI

Abstract

Electroencephalogram (EEG)-based signal processing methods are essential clinical tools to determine and monitor neurological brain disorders such as autism. This article introduces a novel proposal to integrate various neuroimaging methods to characterize an autistic brain. In fact, it is challenging to diagnose and detect the disorder; therefore, it requires the most efficient algorithms for detection. A novel autism identification approach with a combination of VMD+PIE+supervised learning approach is propounded, which can fill the existing gap in the field. The EEG dataset is acquired via the Bonn University and Kaggle database to test the proposed method's performance. Firstly, the VMD technique is used for extracting features from each EEG signal. Then the Predictor Importance Estimates (PIE) have been employed to select the best features from the extracted features. Finally, using supervised learning algorithms (KNN, SVM and ANN), the signals are categorized into a normal or autistic group. The outcome illustrates that the proposed technique attains high accuracy, indicating a powerful way to diagnose and categorize autism.