Electronic Theses and Dissertation

Elektrokardiogram (EKG) adalah salah satu metode yang paling penting dalam
diagnosis penyakit jantung, termasuk deteksi aritmia, yang bertujuan untuk
mengembangkan metode otomatis dalam mendiagnosis aritmia, kondisi di mana
denyut jantung tidak teratur. Aritmia dapat menyebabkan gejala seperti palpitasi,
sesak nafas, dan kelelahan, serta peningkatan risiko stroke dan serangan jantung.
Penelitian ini menggunakan metode Wavelet Scattering Transform (WST) dan BiLSTM (Bidirectional Long-Term-Short Memory) untuk menganalisis sinyal EKG.
Proses ekstraksi fitur dilakukan menggunakan WST untuk mendapatkan
representasi invariant dari translasi lokal yang kemudian diubah menjadi citra
dalam bentuk spektrogram. Sinyal EKG kemudian diklasifikasikan ke dalam tiga
kelas, yaitu Normal, Supraventrikel Takikardia (SVTA), dan Ventrikel Fibrilasi
(VF), menggunakan Bi-LSTM. Hasil evaluasi model menunjukkan bahwa model
yang diusulkan berhasil mencapai akurasi pelatihan sebesar 94,66% dan akurasi
validasi sebesar 98,39%. Selain itu, untuk kelas Normal, model memiliki precision
sebesar 94%, recall sebesar 86%, dan F1-score sebesar 90%. Untuk kelas SVTA,
precision, recall, dan F1-score masing-masing mencapai 99%, 100%, dan 99%.
Sementara itu, untuk kelas VF, precision sebesar 88%, recall sebesar 95%, dan F1-
score sebesar 91%. Hasil ini menunjukkan efektivitas metode yang diusulkan dalam
menganalisis sinyal EKG, yang diharapkan dapat berkontribusi pada
pengembangan metode diagnosis aritmia yang lebih akurat.
Kata Kunci: Elektrokardiogram (EKG), Aritmia, Wavelet Scattering Transform
(WST), Bidirectional Long Short-Term Memory (Bi-LSTM)

Electrocardiogram (ECG) is one of the most important methods in the diagnosis of heart disease, including arrhythmia detection, aiming to develop an automated method of diagnosing arrhythmia, a condition in which the heart rate is irregular. Arrhythmia can cause symptoms such as palpitations, shortness of breath, and fatigue, as well as an increased risk of stroke and heart attack. This research uses Wavelet Scattering Transform (WST) and Bi-LSTM (Bidirectional Long-TermShort Memory) methods to analyze ECG signals. The feature extraction process is performed using WST to obtain an invariant representation of the local translation which is then converted into an image in the form of a spectrogram. ECG signals are then classified into three classes, namely Normal, Supraventricular Tachycardia (SVTA), and Ventricular Fibrillation (VF), using Bi-LSTM. The model evaluation results showed that the proposed model successfully achieved a training accuracy of 94.66% and a validation accuracy of 98.39%. In addition, for the Normal class, the model has a precision of 94%, recall of 86%, and F1-score of 90%. For the SVTA class, the precision, recall, and F1-score reached 99%, 100%, and 99%, respectively. Meanwhile, for the VF class, the precision was 88%, recall was 95%, and F1-score was 91%. These results demonstrate the effectiveness of the proposed method in analyzing ECG signals, which is expected to contribute to the development of more accurate arrhythmia diagnosis methods. Keywords: Electrocardiogram (ECG), Arrhythmia, Wavelet Scattering Transform (WST), Bidirectional Long Short-Term Memory (Bi-LSTM)