Electronic Theses and Dissertation

Abstrak— Autism Spectrum Disorder (ASD) merupakan gangguan neuroperkembangan
yang ditandai dengan hambatan dalam komunikasi, interaksi sosial, serta perilaku repetitif.
Salah satu pendekatan yang banyak digunakan untuk mendukung deteksi ASD adalah
Electroencephalography (EEG) karena mampu merekam aktivitas listrik otak secara real-time
dengan resolusi temporal yang tinggi. Penelitian ini bertujuan untuk menganalisis efektivitas
ekstraksi fitur berbasis Discrete Wavelet Transform (DWT) pada sinyal EEG pita beta (13–
30 Hz) serta membandingkan performa metode Least Squares Support Vector Machine (LS-
SVM), Extreme Gradient Boosting (XGBoost), dan Convolutional Neural Network (CNN)
dalam mengklasifikasikan data ASD dan Normal. Tahap pre-processing dilakukan
menggunakan Independent Component Analysis (ICA) untuk menghilangkan artefak,
dilanjutkan dengan penyaringan pita beta, segmentasi menggunakan overlapping windowing,
dan ekstraksi fitur statistik menggunakan DWT dengan wavelet Daubechies 4 (db4). Dataset
penelitian berasal dari 18 partisipan yang menghasilkan 8.379 epoch setelah proses
segmentasi sinyal EEG. Fitur statistik hasil ekstraksi DWT direpresentasikan menjadi 320
fitur, kemudian data dibagi menjadi data latih dan data uji dengan rasio 90:10. Data latih
dievaluasi menggunakan metode Stratified 9-Fold Cross Validation untuk membandingkan
performa ketiga metode klasifikasi. Hasil penelitian menunjukkan bahwa metode LS-SVM
memberikan performa terbaik dengan nilai accuracy sebesar 97,2%, precision 96,7%, recall
97,1%, specificity 97,7%, dan F1-score 96,9%. Sementara itu, XGBoost memperoleh
accuracy sebesar 90,7% dan F1-score 90,3%, sedangkan CNN memperoleh accuracy sebesar
84,6% dan F1-score 83,1%. Hasil pengujian statistik menggunakan One-Way ANOVA
menunjukkan adanya perbedaan performa yang signifikan antar ketiga metode klasifikasi (p
< 0,05). Dengan demikian, LS-SVM merupakan metode yang paling efektif untuk klasifikasi
sinyal EEG dalam mendeteksi ASD berdasarkan fitur hasil ekstraksi DWT.
Kata kunci: Autism Spectrum Disorder (ASD), Electroencephalography (EEG), Sinyal Beta,
Discrete Wavelet Transform (DWT), LS-SVM, XGBoost, CNN.

Abstract— Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder characterized by impairments in communication, social interaction, and repetitive behaviors. One of the approaches widely used to support ASD detection is Electroencephalography (EEG), as it can record brain electrical activity in real-time with high temporal resolution. This study aims to analyze the effectiveness of feature extraction based on Discrete Wavelet Transform (DWT) on beta-band EEG signals (13–30 Hz) and compare the performance of Least Squares Support Vector Machine (LS-SVM), Extreme Gradient Boosting (XGBoost), and Convolutional Neural Network (CNN) methods in classifying ASD and Normal data. The pre-processing stage was performed using Independent Component Analysis (ICA) to remove artifacts, followed by beta-band filtering, segmentation using overlapping windowing, and statistical feature extraction using DWT with Daubechies 4 (db4) wavelet. The dataset was obtained from 18 participants, producing 8,379 epochs after EEG signal segmentation. The statistical features extracted using DWT were represented into 320 features, and the data were divided into training and testing sets with a 90:10 ratio. The training data were evaluated using Stratified 9-Fold Cross Validation to compare the performance of the three classification methods. The results showed that LS-SVM achieved the best performance with an accuracy of 97.2%, precision of 96.7%, recall of 97.1%, specificity of 97.7%, and F1-score of 96.9%. Meanwhile, XGBoost achieved an accuracy of 90.7% and an F1-score of 90.3%, while CNN obtained an accuracy of 84.6% and an F1-score of 83.1%. Statistical testing using One-Way ANOVA indicated a significant difference in performance among the three classification methods (p < 0.05). Therefore, LS-SVM is the most effective method for EEG signal classification in ASD detection based on DWT-extracted features. Keywords: Autism Spectrum Disorder (ASD), Electroencephalography (EEG), Beta-Band EEG, Discrete Wavelet Transform (DWT), LS-SVM, XGBoost, CNN.