IMPLEMENTASI STATIONARY WAVELET TRANSFORM DAN ARSITEKTUR DENSENET-LONG SHORT-TERM MEMORY PADA KLASIFIKASI BEAT SINYAL ELEKTROKARDIOGRAM

PERTIWI, CITRA and Desiani, Anita and Andriani, Yuli (2025) IMPLEMENTASI STATIONARY WAVELET TRANSFORM DAN ARSITEKTUR DENSENET-LONG SHORT-TERM MEMORY PADA KLASIFIKASI BEAT SINYAL ELEKTROKARDIOGRAM. Undergraduate thesis, Sriwijaya University.

[thumbnail of RAMA_44201_0811382126104_cover.jpg]
Preview
 Image
RAMA_44201_0811382126104_cover.jpg - Cover Image
Available under License Creative Commons Public Domain Dedication.
Download (134kB) | Preview
[thumbnail of RAMA_44201_08011382126104.pdf] Text
RAMA_44201_08011382126104.pdf - Accepted Version
Restricted to Repository staff only
Available under License Creative Commons Public Domain Dedication.
Download (24MB)
[thumbnail of RAMA_44201_08011382126104_TURNITIN.pdf] Text
RAMA_44201_08011382126104_TURNITIN.pdf - Accepted Version
Restricted to Repository staff only
Available under License Creative Commons Public Domain Dedication.
Download (21MB)
[thumbnail of RAMA_44201_08011382126104_0011127702_0002077202_01_front_ref.pdf] Text
RAMA_44201_08011382126104_0011127702_0002077202_01_front_ref.pdf - Accepted Version
Available under License Creative Commons Public Domain Dedication.
Download (6MB)
[thumbnail of RAMA_44201_08011382126104_0011127702_0002077202_02.pdf] Text
RAMA_44201_08011382126104_0011127702_0002077202_02.pdf - Accepted Version
Restricted to Repository staff only
Available under License Creative Commons Public Domain Dedication.
Download (4MB)
[thumbnail of RAMA_44201_08011382126104_0011127702_0002077202_03.pdf] Text
RAMA_44201_08011382126104_0011127702_0002077202_03.pdf - Accepted Version
Restricted to Repository staff only
Available under License Creative Commons Public Domain Dedication.
Download (1MB)
[thumbnail of RAMA_44201_08011382126104_0011127702_0002077202_04.pdf] Text
RAMA_44201_08011382126104_0011127702_0002077202_04.pdf - Accepted Version
Restricted to Repository staff only
Available under License Creative Commons Public Domain Dedication.
Download (11MB)
[thumbnail of RAMA_44201_08011382126104_0011127702_0002077202_05.pdf] Text
RAMA_44201_08011382126104_0011127702_0002077202_05.pdf - Accepted Version
Restricted to Repository staff only
Available under License Creative Commons Public Domain Dedication.
Download (396kB)
[thumbnail of RAMA_44201_08011382126104_0011127702_0002077202_06_ref.pdf] Text
RAMA_44201_08011382126104_0011127702_0002077202_06_ref.pdf - Bibliography
Restricted to Repository staff only
Available under License Creative Commons Public Domain Dedication.
Download (1MB)

Abstract

ECG signal recordings are used to diagnose various heart conditions. Diagnosis of heart abnormalities in ECG signal recordings is done by classifying the heartbeat rhythm into 5 classes, namely non ectopic beat (N), supraventricular ectopic beats (S), ventricular ectopic beats (V), fusion beat (F) and unknown beat (Q). This research will combine the preprocessing stage using Stationary Wavelet Transform (SWT) for signal quality improvement and beat classification of ECG signals in the MIT-BIH Arrhythmia Database using a combination of DenseNet and LSTM architectures. DenseNet is used to capture features from ECG signals through direct connection between layers and LSTM is used to filter features from DenseNet according to data order through gated mechanisms. The SWT method obtained an average SNR value of 26.65 dB, indicating good signal quality with low noise. Model performance evaluation was conducted by measuring accuracy, sensitivity, specificity, precision and F1-score. The results of the performance evaluation obtained an average accuracy of 98.07%, indicating that the model can classify almost all data correctly. The average sensitivity of 95.15% shows that the model can group data that is a certain class correctly. The average specificity of 98.78% shows that the model can group data that is not a certain class correctly. The average precision of 95.17% indicates the model has excellent accuracy in predicting each class. The average F1-score of 95.1% indicates the model is very good at maintaining balance in distinguishing each class, both the class and those that do not belong to the class. The results per class in classes F, S, V, and Q were excellent in all performance evaluation metrics at more than 90%. In class N, the accuracy, specificity, and precision values were very good at more than 90%, but the sensitivity and F1-score were less than 90%. Based on this study, the proposed model provides excellent results overall, but development is needed to distinguish between normal and arrhythmic beats, as well as improvements to the architecture to increase the sensitivity and F1-score values in class N which are less than 90%.

Item Type: Thesis (Undergraduate)
Uncontrolled Keywords: ECG Signal Classification, DenseNet, LSTM, Stationary Wavelet Transform, Signal-to-Noise Ratio
Subjects: Q Science > QA Mathematics > QA8.9-QA10.3 Computer science. Artificial intelligence. Computational complexity. Data structures (Computer scienc. Mathematical Logic and Formal Languages
Divisions: 08-Faculty of Mathematics and Natural Science > 44201-Mathematics (S1)
Depositing User: Citra Pertiwi
Date Deposited: 21 Mar 2025 06:42
Last Modified: 21 Mar 2025 06:42
URI: http://repository.unsri.ac.id/id/eprint/169711

Actions (login required)

View Item View Item