Evaluasi Komparatif Metode Ekstraksi Fitur Domain Waktu pada Sinyal Electroencephalogram untuk Klasifikasi Tingkat Depresi

Azizah, Dyah Ayu Nur (2026) Evaluasi Komparatif Metode Ekstraksi Fitur Domain Waktu pada Sinyal Electroencephalogram untuk Klasifikasi Tingkat Depresi. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5003221039-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (9MB) | Request a copy

Abstract

Depresi merupakan gangguan kesehatan mental yang perlu dideteksi sejak dini karena dapat menurunkan kualitas hidup serta meningkatkan risiko bunuh diri. Salah satu pendekatan yang dapat digunakan untuk mendukung deteksi depresi adalah analisis sinyal electroencephalogram (EEG), yang dapat memberikan informasi fisiologis sebagai pelengkap hasil asesmen psikologis. Penelitian ini bertujuan membandingkan performa empat metode ekstraksi fitur domain waktu, yaitu fitur statistik, parameter Hjorth, Zero Crossing Rate (ZCR), dan Higuchi Fractal Dimension (HFD), serta mengidentifikasi fitur yang paling berkontribusi dalam klasifikasi tingkat depresi berdasarkan sinyal EEG. Data yang digunakan berupa rekaman EEG pada channel Fp1 dan Fp2 dengan kondisi mata tertutup (closed eyes) dari 24 sukarelawan yang terdiri atas 12 individu normal, 9 individu dengan depresi ringan, dan 3 individu dengan depresi sedang. Sinyal EEG diproses melalui tahapan preprocessing, filtrasi ke dalam lima subband frekuensi, segmentasi, dan ekstraksi fitur. Seleksi fitur dilakukan menggunakan uji Kruskal–Wallis, sedangkan proses klasifikasi menggunakan Random Forest dengan validasi Stratified Group K-Fold Cross Validation. Hasil penelitian menunjukkan bahwa fitur statistik memberikan performa klasifikasi terbaik dibandingkan metode ekstraksi fitur lainnya dengan testing accuracy sebesar 62,24%, macro precision sebesar 51,60%, macro recall sebesar 47,44%, macro F1-score sebesar 46,11%, dan macro AUC sebesar 72,47%. Analisis feature selection dan feature importance menunjukkan bahwa fitur berbasis varians merupakan fitur domain waktu yang paling berkontribusi dalam membedakan tingkat depresi. Namun, model masih menunjukkan keterbatasan dalam mengidentifikasi kelas depresi sedang, sehingga pengembangan metode klasifikasi masih diperlukan untuk meningkatkan kemampuan diskriminasi terhadap kelas tersebut.
=================================================================================================================================
Depression is a mental health disorder that requires early detection because it can reduce quality of life and increase the risk of suicide. One approach that can be used to support depression detection is electroencephalogram (EEG) signal analysis, which provides physiological information as a complement to psychological assessments. This study aims to compare the performance of four time-domain feature extraction methods, namely statistical features, Hjorth parameters, Zero Crossing Rate (ZCR), and Higuchi Fractal Dimension (HFD), and to identify the features that contribute most significantly to the classification of depression levels based on EEG signals. The data used in this study consisted of EEG recordings from the Fp1 and Fp2 channels under a closed-eyes condition obtained from 24 volunteers, including 12 normal individuals, 9 individuals with mild depression, and 3 individuals with moderate depression. The EEG signals were processed through preprocessing, filtering into five frequency subbands, segmentation, and feature extraction stages. Feature selection was performed using the Kruskal–Wallis test, while classification was conducted using Random Forest with Stratified Group K-Fold Cross Validation. The results showed that statistical features achieved the best classification performance compared with other feature extraction methods, with a testing accuracy of 62.24%, macro precision of 51.60%, macro recall of 47.44%, macro F1-score of 46.11%, and macro AUC of 72.47%. Feature selection and feature importance analyses indicated that variance-based features were the most influential time-domain features in distinguishing depression levels. However, the model still showed limitations in identifying the moderate depression class, indicating that further development of classification methods is required to improve discrimination capability for this class.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Depresi, Domain Waktu, Electroencephalogram, Klasifikasi, Random Forest, Depression, Time Domain, Electroencephalogram, Classification, Random Forest
Subjects:	Q Science > QP Physiology > Q376.5 Electroencephalography (EEG)
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49201-(S1) Undergraduate Thesis
Depositing User:	Dyah Ayu Nur Azizah
Date Deposited:	01 Aug 2026 03:34
Last Modified:	01 Aug 2026 03:34
URI:	http://repository.its.ac.id/id/eprint/141736

Actions (login required)

View Item