Beetle Swarm-based Multi Verse Optimization for the Autism Spectrum Disorder Detection via EEG Signal Analysis: A Novel Hybrid Deep Learning Concept

Abstract

Traditional automated EEG-oriented ASD diagnosis utilizing several nonlinear EEG analysis approaches was restricted to distinguishing just children having ASD from those who were typically developing without reaching the severity of their autistic traits. Using EEG research to identify possible variations among children having mild and severe ASD is a serious challenge. Hence, this paper plans to detect the ASD via the EEG signal processing and analysis. The dataset is first gathered from the ASD dataset. After that, median filtering and normalization are used to complete the pre-processing. The signals are then transformed to images using the STFT method. GLCM is used to extract features from the transformed images. Because the retrieved features are long, PCA is used to pick significant features. The categorization is done using a hybrid deep learning model based on these ideally picked features. SVM takes the place of CNN's fully connected layer in this case. Novel BS-MVO by merging BSO and MVO optimizes the learning rate in CNN and iteration in SVM with the goal of accuracy maximization as the fitness or objective function, resulting in enhanced CNN. The final output is classified using this enhanced CNN. The accuracy of the proposed BS-MVO at 90% learning percentage is higher with 94.49% in comparison to GA with 92.90%, WOA with 89.99%, MVO with 93.91%, and BSO with 91.60% respectively.  Simulation findings demonstrate the superiority of the proposed method by comparing with traditional methods in terms of various measures.

DOI: 10.61416/ceai.v27i3.9570