Electronic Theses and Dissertation

Abstrak-Autism Spectrum Disorder (ASD) adalah gangguan perkembangan
neurologis yang berdampak pada komunikasi, perilaku, dan interaksi sosial.
Penelitian ini bertujuan mengklasifikasikan sinyal EEG beta (12–30 Hz) dari anak
ASD dan anak normal menggunakan metode Discrete Wavelet Transform (DWT)
dan algoritma Least Squares Support Vector Machine (LS-SVM). Data EEG dari
16 subjek dipra-proses dengan Independent Component Analysis (ICA), lalu
disegmentasi menggunakan overlapping window berdurasi 2 detik dengan 50%
overlap, menghasilkan 8.379 data berdimensi 16 channel × 512 sampel. Fitur
diekstraksi menggunakan DWT (db4, 4 level) menghasilkan 320 fitur statistik,
kemudian dinormalisasi menggunakan MinMaxScaler. Data dibagi menjadi 70%
pelatihan dan 30% pengujian. Klasifikasi dilakukan menggunakan LS-SVM dengan
kernel linear dan polynomial. Kernel polynomial menghasilkan akurasi 98,49%,
precision 99,05%, recall 98,20%, dan F1-score 98,62%. Sementara kernel linear
memperoleh akurasi 95,07%, precision 94,07%, recall 97,20%, dan F1-score
95,61%. Hasil ini menunjukkan bahwa kombinasi DWT dan LS-SVM, khususnya
dengan kernel polynomial, sangat efektif dalam membedakan sinyal EEG beta
antara anak ASD dan anak normal.
Kata kunci : Autism Spectrum Disorder (ASD), Electroecephalogram (EEG),
Sinyal Beta, Discrete Wavelet Transform (DWT), Least Squares Support Vector
Machine (LS-SVM).

Abstract – Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder that affects communication, behavior, and social interaction. This study aims to classify beta EEG signals (12–30 Hz) from children with ASD and normal children using the Discrete Wavelet Transform (DWT) method and the Least Squares Support Vector Machine (LS-SVM) algorithm. EEG data from 16 subjects were preprocessed using Independent Component Analysis (ICA), then segmented using 2-second overlapping windows with 50% overlap, resulting in 8,379 samples with dimensions of 16 channels × 512 data points. Features were extracted using DWT (db4, 4 levels), producing 320 statistical features, and then normalized using MinMaxScaler. The dataset was divided into 70% for training and 30% for testing. Classification was performed using LS-SVM with linear and polynomial kernels. The polynomial kernel achieved an accuracy of 98.49%, precision of 99.05%, recall of 98.20%, and an F1-score of 98.62%. Meanwhile, the linear kernel achieved an accuracy of 95.07%, precision of 94.07%, recall of 97.20%, and an F1-score of 95.61%. These results indicate that the combination of DWT and LS-SVM, particularly with the polynomial kernel, is highly effective in distinguishing beta EEG signals between children with ASD and normal children. Keywords: Autism Spectrum Disorder (ASD), Electroencephalogram (EEG), Beta Signals, Discrete Wavelet Transform (DWT), Least Squares Support Vector Machine (LS-SVM).