Muhammad Mirza Rahmat, Yunidar Yunidar, Melinda Melinda
Electroencephalography (EEG) signals are highly susceptible to artifacts, making effective denoising crucial for reliable Autism Spectrum Disorder (ASD) classification. This study presents a controlled comparison between Multi-Scale Principal Component Analysis (MSPCA) and Multi-Scale Independent Component Analysis (MSICA) as multiscale denoising techniques for EEG-based ASD classification. EEG data were acquired from 16 subjects using a 16-channel OpenBCI system and preprocessed with a 0.5-40 Hz Butterworth band-pass filter. Both denoising methods were integrated with Light Gradient Boosting Machine (LightGBM) and a Temporal Convolutional Network with Self-Attention (TCN-SA) under an identical subject-wise cross-validation framework. Experimental results show that LightGBM combined with MSPCA achieved the highest average accuracy of 99.14% and an AUC of 99.96%. Subject-level paired analysis based on ASD posterior probabilities for the ASD subjects showed positive MSPCA–MSICA differences for both LightGBM and TCN-SA. This pattern was supported by the paired t-test, Wilcoxon signed-rank test, and exact paired permutation test, while effect sizes and confidence intervals indicated that the observed advantage should be interpreted as a within-dataset finding given the limited cohort size. This article is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License. License details: https://creativecommons.org/licenses/by-sa/4.0/
Department of Electrical and Computer Engineering, Faculty of Engineering, Universitas Syiah Kuala, Indonesia
Research at a Glance
Register to unlock
Topics & SDG Alignment
Register to unlock
Collaboration
Register to unlock
Author Profile (Selected)
Register to unlock
References Overview
Register to unlock
Journal & Source
Register to unlock
Metadata & Integrity
Register to unlock