The Impact of Multiple Comorbidities on ST-Segment Abnormalities in Electrocardiogram (ECG) Signals with Machine Learning

Abstract

 In this work, a rigorous comparative analysis of five machine learning to choose which of them best ensemble models  are AdaBoost, GB, CatBoost, XGBoost, and LightGB, was performed on four varying data partitioning schemes (from 70-30 to 85-15 train-test splits) for maximizing clinical risk stratification. The sample consisted of (1000) ST-segment abnormal patients, received from Shar Hospital in Sulaymaniyah City's. The dataset included age, gender, smoking status, hypertension, chronic cardiovascular disease, diabetes mellitus, dyslipidemia, and hypothyroidism. Chi-square test revealed substantial correlations among the predictor variables. The five boosting models were subsequently trained and tested under a range of data-splitting regimes, with performance metrics in terms of accuracy, F1-score, recall, and precision. Computational protocols were executed by using Python programming Results: by using Chi-square test get significant correlation between ST-segment and each of Age, Hypertension, Chronic Cardiovascular and Dyslipidemia. After that best Algorithm is AdaBoost performed best discriminatively, with a highly good F1-score of (0.8839) and maximum flawless recall (1.000) at (75-25) train-test split in all experiments. Feature Importance: Age was the strongest predictor, which reaffirms its prognostic value in ST-segment related pathology.