Penerapan Independent Component Analysis untuk Ekstraksi Fitur pada Klasifikasi Aritmia Jantung Berbasis Light Gradient Boosting Machine

Abstract

Cardiovascular disease, specifically arrhythmia, is a major health threat and one of the leading cause of death worldwide. Manually monitoring and detecting arrhythmias using an Electrocardiogram (ECG) is prone to human error due to the visual fatigue of medical staff. This research proposes a machine learning-based system to assist healthcare professionals in classifying arrhythmias more efficiently. The data used were obtained from the MIT-BIH Arrhythmia Database, focusing on the Modified Limb Lead II (MLII) channel. Data preprocessing was performed using cascading filters (High-pass, Low-pass, and Notch filters) and Z-score normalization to reduce noise, followed by beat segmentation based on three AAMI classes (N,S, and V). Principal Component Analysis (PCA) was applied prior to Independent Component Analysis (ICA) to reduce dimensionality. The primary focus of this research is the comparison of two ICA algorithms, namely FastICA and Picard, combined with morphological, statistical, and RR-interval features for ECG representation. To address class imbalance, sample weights were calculated from the training data and incorporated into the classification model. Classification was performed using Light Gradient Boosting Machine (LightGBM). The results show that the proposed feature combination and LightGBM model achieved good performance for arrhythmia classification on real-world data, with FastICA achived a macro F1-score of 0.7238, outperforming Picard which achieved 0.6963.