Perancangan Model Untuk Mendeteksi Penyakit Kardiovaskular Berbasis SVM Dan Teknik Penyeimbangan Data Berbasis Fitur Hibrida Pada Sinyal EKG PTB-XL

Hidayat, Tsaabitah Rizqiina Putri (2026) Perancangan Model Untuk Mendeteksi Penyakit Kardiovaskular Berbasis SVM Dan Teknik Penyeimbangan Data Berbasis Fitur Hibrida Pada Sinyal EKG PTB-XL. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6026251027-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (10MB) | Request a copy

Abstract

Ketidakseimbangan kelas pada dataset EKG 12-lead PTB-XL menurunkan sensitivitas klasifikasi penyakit kardiak langka seperti Hypertrophy (HYP). Penelitian ini mengomparasikan efektivitas pendekatan Data-Centric (SMOTE) dan Algorithm-Centric (Weighted BCE) dalam mengatasi bias kelas mayoritas pada klasifikasi 5 superclass EKG. Tahap pemrosesan awal menerapkan pemfilteran Butterworth, IIR notch, DWT db4, dan normalisasi Z-score. Ekstraksi fitur hibrida mencakup morfologi P-QRS-T, polinomial Hermite, dan korelasi Spearman inter-lead. Seleksi fitur dua tahap (mRMR dan Boruta) serta PCA menghasilkan 61 komponen utama bebas multikolinearitas yang menjelaskan 95,15% varians kumulatif. Klasifikasi dievaluasi menggunakan Logistic Regression, Random Forest, dan SVM dengan optimasi hiperparameter Optuna berbasis Official 10-Fold Split. Hasil eksperimen menunjukkan bahwa Weighted BCE mengungguli SMOTE. Model SVM dengan Weighted BCE mencapai performa puncak dengan Macro-F1 0,6861, Macro-Recall 0,7019, dan Akurasi Global 0,7879, serta meningkatkan Recall kelas minoritas HYP hingga 55,36% pada Logistic Regression.
======================================================================================================================================
Class imbalance in the 12-lead PTB-XL ECG dataset degrades classification sensitivity for rare cardiac conditions such as Hypertrophy (HYP). This study compares Data-Centric (SMOTE) and Algorithm-Centric (Weighted BCE) approaches to address majority class bias in 5-superclass ECG classification. Preprocessing incorporates Butterworth filtering, IIR notch, DWT db4 denoising, and Z-score normalization. Hybrid feature extraction captures P-QRS-T morphology, Hermite polynomials, and inter-lead Spearman correlations. A two-stage feature selection (mRMR and Boruta) combined with PCA yields 61 orthogonal principal components explaining 95.15% cumulative variance. Classification is evaluated using Logistic Regression, Random Forest, and SVM optimized via Optuna under the official 10-fold split. Results demonstrate that Weighted BCE superiorly mitigates class imbalance compared to SMOTE. The SVM model paired with Weighted BCE achieves optimal performance with a Macro-F1 of 0.6861, Macro-Recall of 0.7019, and Global Accuracy of 0.7879, significantly increasing HYP minority class recall to 55.36% in Logistic Regression.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Elektrokardiogram, Keseimbangan Data, SMOTE, Weighted BCE Machine Learning, PTB-XL, Electrocardiogram, Data Imbalance, SMOTE, Weighted BCE, Machine Learning, PTB-XL
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information System > 59101-(S2) Master Thesis
Depositing User:	Tsaabitah Rizqiina Putri Hidayat
Date Deposited:	03 Aug 2026 04:25
Last Modified:	03 Aug 2026 04:25
URI:	http://repository.its.ac.id/id/eprint/142504

Actions (login required)

View Item