Deteksi Diabetes Melitus Tipe 2 Menggunakan Sinyal Fotopletismografi Dan Elektrokardiografi Berbasis Machine Learning

Ilmiati, Dewi Martha (2025) Deteksi Diabetes Melitus Tipe 2 Menggunakan Sinyal Fotopletismografi Dan Elektrokardiografi Berbasis Machine Learning. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5023211037-Undergraduated_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB)

Abstract

Ketergantungan terhadap metoded diagnostik invasif menjadi tantangan dalam deteksi diabetes mellitus tipe 2 (DMT2), sehingga kenyamanan pasien terganggu. Peelitian ini mengusulkan metode deteksi risiko diabetes mellitus tipe 2 secara non-invasif dengan mengintegrasikan sinyal Photoplethysmography (PPG) dan Elektrokardiogram (ECG). Integrasi kedua sinyal ini memungkinkan ekstrasi fitur-fitur yang berkatan erat dengan gula darah, seperti Pulse Transit Time (PTT) yang terkait dengan kekakuan arteri, serta fitur-fitur domain waktu lainnya. Selanjutnya, dikembangkan model machine learning XGBoost untuk memodelkan hubungan kompleks antara fitur-fitur yang terpilih menggunakan metode seleksi Recursive Feature Elimination. Validasi model dilakukan mengguanakan Leave One Out Cross Validation (LOOCV) untuk mendapatkan estimasi performa yang reliable pada tiap data. Dari model yang telah diterapkan telah didapatkan R-squared (R²) sebesar 0.4729 dan Mean Absolute Error (MAE) untuk prediksi kadar gula darah sebesar 13,12 mg/dLdan Root Mean Square Error (RMSE) sebesar 17.60 mg/dL. Tahapan selanjutnya, ketika hasil estimasi digunakan untuk klasifikasi subjek menggunakan threshold kadar gula darah yang berlaku, model menunjukkan kemampuan yang baik dalam membedakan antara subjek dengan kondisi normal dan diabetes. Secara keseluruhan, meskipun akurasi dari model masih memiliki ruang untuk perbaikan teteapi pendekatan ini memiliki potensi signifikan sebagaia alat skrining non-invasif untuk deteksi dini risiko diabetes mellitus tipe 2.
==========================================================================================================================================
Dependence on invasive diagnostic methods poses a challenge in the detection of type 2 diabetes mellitus (T2DM), compromising patient comfort. This research proposes a non-invasive method for detecting T2DM risk by integrating Photoplethysmography (PPG) and Electrocardiogram (ECG) signals. The integration of these two signals allows for the extraction of features closely related to blood glucose, such as Pulse Transit Time (PTT), which is associated with arterial stiffness, as well as other time-domain features. Furthermore, a XGBoost machine learning model was developed to model the complex relationship between features selected using Recursive Feature Elimination ccorrelation methods. Model validation was performed using Leave-One-Out Cross-Validation (LOOCV) to obtain a reliable performance estimate. The implemented model achieved an R-squared (R²) of 0.4729, a Mean Absolute Error (MAE) of 13,12 mg/dL, and a Root Mean Square Error (RMSE) of 17.60 mg/dL for blood glucose prediction. Subsequently, when the estimation results were used to classify subjects based on established blood glucose thresholds, the model demonstrated a good ability to distinguish between normal and diabetic conditions. Overall, although the model's accuracy has room for improvement, this approach holds significant potential as a non-invasive screening tool for the early detection of type 2 diabetes mellitus risk.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Type 2 Diabetes Mellitus, non-invasive, Photoplethysmography (PPG), Electrocardiogram (ECG), blood glucose.
Subjects:	T Technology > T Technology (General) > T57.74 Linear programming T Technology > T Technology (General) > T58.62 Decision support systems
Divisions:	Faculty of Electrical Technology > Biomedical Engineering > 11410-(S1) Undergraduate Thesis
Depositing User:	Dewi Martha Ilmiati
Date Deposited:	05 Aug 2025 07:01
Last Modified:	05 Aug 2025 07:03
URI:	http://repository.its.ac.id/id/eprint/127532

Actions (login required)

View Item