Deteksi Kondisi Relaks melalui Seleksi Fitur Spasio Temporal untuk Sub-band Alpha pada Data EEG

Risqiwati, Diah (2024) Deteksi Kondisi Relaks melalui Seleksi Fitur Spasio Temporal untuk Sub-band Alpha pada Data EEG. Doctoral thesis, Institut Teknologi Sepuluh Nopember.

Text 07111960010012-Dissertation.pdf - Accepted Version Restricted to Repository staff only until 1 April 2027. Download (3MB) | Request a copy

Abstract

Relaks adalah bagian dari emosi manusia yang dapat dipetakan dalam kuadran dua dimensi arousal-valence. Arousal menunjukkan intensitas emosi (bergairah/tidak bergairah), sedangkan valence menunjukkan polaritas (positif/negatif). Emosi dibagi menjadi empat kategori yaitu: sedih (arousal rendah, valence rendah), relaks (arousal rendah, valence tinggi), stres (arousal tinggi, valence rendah), dan bahagia (arousal tinggi, valence tinggi). Relaks sangat penting dieksplorasi karena bermanfaat sebagai terapi stres dan gangguan mental. Kondisi relaks dapat diukur secara fisiologis, seperti melalui kerenggangan otot, kadar oksigen, tekanan darah, dan denyut jantung yang normal. Namun variasi kondisi fisik antar individu sering memunculkan ketidakpastian. Untuk mengatasi ini, EEG digunakan karena mampu memberikan gambaran akurat tentang aktivitas otak dan emosi seseorang. Dalam terapi hipnosis, deteksi kondisi relaks sangat penting, karena keberhasilan pemberian sugesti bergantung pada tercapainya kondisi ini. Dari penelitian sebelumnya didapatkan bahwa sinyal yang paling dominan dalam pengamatan kondisi relaks adalah alpha band. Oleh karena itu penelitian ini memfokuskan pengamatan pada alpha band dengan memisahkan sub-band alpha tinggi dan alpha rendah. Belum ada fitur spesifik yang dapat menggambarkan karakteristik sinyal alpha untuk kondisi relaks secara optimal. Transformasi wavelet menggunakan Continuous Wavelet Transform (CWT), telah terbukti unggul dalam menganalisis sinyal alpha yang bersifat non-stasioner dalam mengenali kondisi relaks dibandingkan metode lain seperti Empirical Mode Decomposition (EMD) dan Discrete Wavelet Transform (DWT). Penelitian ini mengeksplorasi 13 fitur yang umum digunakan dalam pengenalan emosi. Pada metode Effective Relax Acquisition (ERA), dilakukan seleksi dan reduksi fitur menggunakan Principal Component Analysis (PCA), menghasilkan 6 fitur signifikan untuk mendeteksi kondisi relaks. Fiturfitur ini kemudian dimasukkan ke dalam classifier, dengan akurasi tertinggi diperoleh menggunakan K-Nearest Neighbour (K-NN), yaitu sebesar 90,63%. Hasil ini dicapai di sub-band alpha tinggi pada domain frekuensi dan di periode grup ke-4 pada domain waktu. Dalam percobaan berikutnya, metode reduksi fitur Relief-F menghasilkan dua fitur spasio-temporal terbaik, yaitu amplitudo maksimum dan standar deviasi. Kedua fitur ini digunakan untuk menyusun rules Fuzzy Intuitionistic dan Mamdani, dengan metode yang disebut Mamdani Intuitionistic Fuzzy Rules Set (MIFRS). Penggunaan MIFRS meningkatkan akurasi prediksi kondisi relaks menjadi 92,45%, baik pada subband alpha tinggi di domain frekuensi dan pada beberapa segmen waktu dalam jendela segmen overlapping shifting di domain waktu. Prediksi kondisi relaks yang dihasilkan telah divalidasi oleh tiga neurologis dan menunjukkan bahwa MIFRS mampu mendeteksi onset kondisi relaks dengan lebih akurat dengan menggunakan jumlah fitur spasio-temporal yang lebih sedikit dibandingkan metode ERA.
==================================================================================================================================
Relaxed state is a subset of human emotions that can be mapped in a two-dimensional quadrant. Arousal indicates the intensity of the emotion (excited/not excited), while valence indicates the polarity (positive/negative). Emotions are divided into four categories: sad (low arousal, low valence), relaxed (low arousal, high valence), stressed (high arousal, low valence), and happy (high arousal, high valence). Relaxation is very important to explore as it is useful as a therapy for stress and mental disorders. Relaxed states can be measured physiologically, such as through normal muscle tone, oxygen levels, blood pressure and heart rate. However, variations in physical conditions between individuals often lead to uncertainty. To overcome this, EEG is used because it is able to provide an accurate picture of a person’s brain activity and emotions. In hypnosis therapy, the detection of a relaxed state is very important, as the success of suggestion depends on achieving this state. From previous research, it was found that the most dominant signal in the observation of relaxed state is alpha band. Therefore, this study focuses on the alpha band by separating the high alpha and low alpha sub-bands. There is no specific feature that can optimally describe the characteristics of the alpha signal for relaxed state. Wavelet transformation using Continuous Wavelet Transform (CWT), has proven superior in analyzing non-stationary alpha signals in recognizing relaxed states compared to other methods such as Empirical Mode Decomposition (EMD) and Discrete Wavelet Transform (DWT). This research explores 13 features that are commonly used in emotion recognition. In the Effective Relax Acquisition (ERA) method, feature selection and reduction using Principal Component Analysis (PCA) was performed, resulting in 6 significant features for detecting relaxed states. These features were then incorporated into a classifier, with the highest accuracy obtained using K-Nearest Neighbor (K-NN), which was 90.63%. This result was achieved in the high alpha sub-band in the frequency domain and in the 4th group period in the time domain. In the next experiment, the Relief-F feature reduction method produced the two best spatio-temporal features, namely maximum amplitude and standard deviation. These two features were used to construct Intuitionistic and Mamdani fuzzy rules, in a method called Mamdani Intuitionistic Fuzzy Rules Set (MIFRS). The use of MIFRS increased the accuracy of relaxed state prediction to 92.45%, both at the high alpha sub-band in the frequency domain and at multiple time segments within the overlapping shifting segment window in the time domain. The resulting relaxed state prediction has been validated by three neurologists and shows that MIFRS is able to detect relaxed state onset more accurately using a smaller number of spatio-temporal features than the ERA method.

Item Type:	Thesis (Doctoral)
Uncontrolled Keywords:	Relaxed state detection, wavelet transform, spatio temporal features, feature selection, feature reduction, PCA, relief-F, intuitionistic fuzzy, mamdani fuzzy, Deteksi kondisi relaks, transformasi wavelet, fitur spasio temporal, seleksi fitur, reduksi fitur
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5102.9 Signal processing.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20001-(S3) PhD Thesis
Depositing User:	Diah Risqiwati
Date Deposited:	09 Jan 2025 08:27
Last Modified:	09 Jan 2025 08:27
URI:	http://repository.its.ac.id/id/eprint/116235

Actions (login required)

View Item