ABSTRACT Autism Spectrum Disorder (ASD) is a neurological condition that affects the cerebrospinal nervous system, impairs motor function, and other developmental aspects. In this study, we proposed a novel approach that ensembled Fractal Dimension (FD) measures with a Bayesian Optimized Artificial Neural Network (BO‐ANN) for the automated classification of multichannel Electroencephalography (EEG) signals into ASD ( n = 29) and Typically Developing (TD) ( n = 30) participants. We extracted five FD measures: Higuchi's Fractal Dimension (HFD), Correlation Dimension (CRD), Box‐Counting Method (BCM), Katz's Fractal Dimension (KFD), and Petrosian's Fractal Dimension (PFD) from EEG signals. These features are assembled into an 80‐dimensional vector (5 FDs × 16 channels) and demonstrate a statistically significant difference between TD and ASD controls (α < 0.05). This feature set subsequently fed into six Machine Learning (ML) classifiers, an ANN model, and the proposed BO‐ANN model for the identification process. Extreme Gradient Boosting (XGB) achieved an accuracy of 90.09%, while the baseline ANN obtained 91.44%. The proposed BO‐ANN outperformed all models with a classification accuracy of 97.38%. Feature importance analysis with XGB further revealed the most influential channels and FD measures contributing to discrimination. Our study provides a robust framework for ASD identification using EEG signals and the proposed BO‐ANN model, emphasizing the effectiveness of FD‐based features. The proposed framework not only improves accuracy compared to the existing approaches but also enhances interpretability, making it a promising tool for practical applications in ASD identification.
Ranaut et al. (Sun,) studied this question.