Penerapan deep learning dalam klasifikasi sinyal ECG: perbandingan metode CNN, LSTM, dan hybrid

Abidin, Muhammad Farid (2024) Penerapan deep learning dalam klasifikasi sinyal ECG: perbandingan metode CNN, LSTM, dan hybrid. Masters thesis, Universitas Pelita Harapan.

[thumbnail of Title] Text (Title)
Title_MUHAMMAD FARID ABIDI.pdf
Available under License Creative Commons Attribution Non-commercial Share Alike.

Download (70kB)
[thumbnail of Abstract] Text (Abstract)
Abstract_MUHAMMAD FARID ABIDI.pdf
Available under License Creative Commons Attribution Non-commercial Share Alike.

Download (192kB)
[thumbnail of ToC] Text (ToC)
ToC_MUHAMMAD FARID ABIDI.pdf
Available under License Creative Commons Attribution Non-commercial Share Alike.

Download (449kB)
[thumbnail of Chapter 1] Text (Chapter 1)
Chapter1_MUHAMMAD FARID ABIDI.pdf
Available under License Creative Commons Attribution Non-commercial Share Alike.

Download (803kB)
[thumbnail of Chapter 2] Text (Chapter 2)
Chapter2_MUHAMMAD FARID ABIDI.pdf
Restricted to Registered users only
Available under License Creative Commons Attribution Non-commercial Share Alike.

Download (734kB)
[thumbnail of Chapter 3] Text (Chapter 3)
Chapter3_MUHAMMAD FARID ABIDI.pdf
Restricted to Registered users only
Available under License Creative Commons Attribution Non-commercial Share Alike.

Download (2MB)
[thumbnail of Chapter 4] Text (Chapter 4)
Chapter4_MUHAMMAD FARID ABIDI.pdf
Restricted to Registered users only
Available under License Creative Commons Attribution Non-commercial Share Alike.

Download (3MB)
[thumbnail of Chapter 5] Text (Chapter 5)
Chapter5_MUHAMMAD FARID ABIDI.pdf
Restricted to Registered users only
Available under License Creative Commons Attribution Non-commercial Share Alike.

Download (368kB)
[thumbnail of Bibliography] Text (Bibliography)
Bibliography_MUHAMMAD FARID ABIDI.pdf
Available under License Creative Commons Attribution Non-commercial Share Alike.

Download (713kB)
[thumbnail of Appendices] Text (Appendices)
Appendices_MUHAMMAD FARID ABIDI.pdf
Restricted to Repository staff only
Available under License Creative Commons Attribution Non-commercial Share Alike.

Download (4MB)

Abstract

Cardiovascular Diseases (CVDs) are one of the leading causes of death globally, responsible for 17.9 million deaths annually or approximately 32% of total global deaths. To address this issue, an early diagnosis of potential CVD in an individual is highly needed. Electrocardiogram (ECG) is a critical diagnostic tool for analyzing the heart's electrical activity. However, manual analysis of ECG signals often requires significant expertise and time, especially when dealing with large volumes of data or complex cases. With the advancement of technology, Machine Learning and Deep Learning applications in ECG signal analysis have rapidly evolved. These methods provide faster and more accurate analysis processes compared to manual approaches. In this study three distinct approaches for ECG signal classification: Convolutional Neural Networks, Long Short-Term Memory, and a Hybrid CNN-LSTM models are designed and compared. Utilizing MATLAB's Deep Network Designer, three relatively simple network architectures are developed for this purpose. To ensure a rigorous comparison, experiments are conducted by varying solvers, number of epoch, and number of segmentation, aiming to identify optimal configurations and maximize classification performance. The results show that the CNN model using Scalograms from Continuous Wavelet Transform (CWT) showed good performance with an accuracy of 91.30% and an F1 Score of 86.67%. However, it struggled with detecting the Congestive Heart Failure (CHF) class. The LSTM model, while effective at capturing temporal patterns, exhibited lower performance with an accuracy of 61.89% and an F1 Score of 62.88%, particularly for CHF and Normal Sinus Rhythm (NSR) classes. The Hybrid approach provided the best results, with an accuracy of 82.60% and an F1 Score of 93.75%, combining CNN's strengths in spatial feature extraction and LSTM's ability to capture temporal patterns. However, CHF detection remains a challenge. The hybrid model shows great potential in ECG classification, although further improvements such as data augmentation, enhanced preprocessing, and more complex architectures are needed to improve performance, particularly for minority classes. / Penyakit jantung adalah salah satu penyebab utama kematian di dunia, bertanggung jawab atas 17,9 juta kematian setiap tahun atau sekitar 32% dari total kematian global. Untuk mengatasi masalah ini, diagnosis dini terhadap potensi CVD pada individu sangat dibutuhkan. Electrocardiogram (ECG) merupakan alat diagnostik penting untuk menganalisis aktivitas listrik jantung. Namun, analisis manual sinyal ECG sering kali membutuhkan keahlian yang tinggi dan waktu yang cukup lama, terutama jika berhadapan dengan volume data yang besar atau kasus yang kompleks. Dengan kemajuan teknologi, aplikasi Pembelajaran Mesin (Machine Learning) dan pembelajaran mendalam (Deep Learning) dalam analisis sinyal ECG berkembang dengan pesat. Metode-metode ini menawarkan analisis yang lebih cepat dan akurat dibandingkan pendekatan manual. Dalam penelitian ini, digunakan tiga pendekatan untuk melakukan klasifikasi sinyal ECG: model Convolutional Neural Networks, Long Short-Term Memory, dan model Hybrid CNN-LSTM. Dengan memanfaatkan Deep Network Designer pada MATLAB, tiga arsitektur jaringan yang relatif sederhana dikembangkan untuk tujuan ini. Untuk memastikan perbandingan yang tepat, eksperimen dilakukan dengan memvariasikan solver, jumlah epoch, dan jumlah segmentasi, guna mengidentifikasi konfigurasi optimal dan memaksimalkan kinerja klasifikasi. Hasil penelitian menunjukkan bahwa model CNN menggunakan Scalogram dari Continuous Wavelet Transform (CWT) memberikan kinerja yang baik dengan akurasi sebesar 91,30% dan F1-score sebesar 86,67%. Namun, model ini mengalami kesulitan dalam mendeteksi kelas Congestive Heart Failure (CHF). Model LSTM, meskipun efektif dalam menangkap pola temporal, menunjukkan kinerja lebih rendah dengan akurasi 61,89% dan F1-score sebesar 62,88%, terutama untuk kelas CHF dan Normal Sinus Rhythm (NSR). Pendekatan Hybrid memberikan hasil terbaik dengan akurasi 82,60% dan F1-score sebesar 93,75%, yang menggabungkan keunggulan CNN dalam ekstraksi fitur spasial dan kemampuan LSTM untuk menangkap pola temporal. Namun, deteksi CHF tetap menjadi tantangan. Model Hybrid menunjukkan potensi besar dalam klasifikasi ECG, meskipun diperlukan optimasi lebih lanjut seperti augmentasi data, prapemrosesan yang lebih baik, dan arsitektur yang lebih kompleks untuk meningkatkan kinerja, terutama untuk kelas minoritas.
Item Type: Thesis (Masters)
Creators:
Creators
NIM
Email
ORCID
Abidin, Muhammad Farid
NIM01679230002
UNSPECIFIED
UNSPECIFIED
Contributors:
Contribution
Contributors
NIDN/NIDK
Email
Thesis advisor
Tjahyadi, Hendra
NIDN0410076901
hendra.tjahyadi@uph.edu
Uncontrolled Keywords: Cardiovascular Disease ; Electrocardiogram ; Deep Learning ; Convolutional Neural Networks ; Long Short Term Memory ; Hybrid CNN-LSTM ; Continuous Wavelet Transform
Subjects: T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7885-7895 Computer engineering. Computer hardware
Divisions: University Subject > Current > Faculty/School - UPH Karawaci > School of Information Science and Technology > Master of Informatics
Current > Faculty/School - UPH Karawaci > School of Information Science and Technology > Master of Informatics
Depositing User: Phillips Iman Heri Wahyudi
Date Deposited: 22 Feb 2025 07:10
Last Modified: 22 Feb 2025 07:10
URI: http://repository.uph.edu/id/eprint/67183

Actions (login required)

View Item
View Item