An XGBoost-based machine learning pipeline achieved 98.01% accuracy and an F1-score of 0.977 for arrhythmia detection using 2,759 heartbeats from the MIT-BIH database.
An XGBoost-based machine learning framework can accurately detect arrhythmias from ECG signals with 98.01% accuracy.
Abstract: All cardiovascular diseases are still the main cause of mortality in the whole world, and arrhythmiasare especially problematic in their diagnostics. This paper presents an end-to-end machine learning system forarrhythmia detection using the MIT-BIH Arrhythmia Database. The proposed pipeline includes: Signal preprocessing (0.5-40 Hz bandpass filtering, median baseline correction, Z-score normalization) Feature extraction (27 morphological, statistical, and spectral features per heartbeat) Dimensionality reduction (PCA, Mutual Information, Recursive Feature Elimination) Systematic comparison of five classifiers: Logistic Regression, SVM, Random Forest, XGBoost, and NeuralNetwork Key Results: XGBoost achieved the highest accuracy (98.01%) and F1-score (0.977) Logistic Regression offered the fastest training time (7.43 seconds) Feature selection demonstrated that 27 features are necessary to maintain discriminative power PCA reduced feature dimension from 250 to 16 while retaining 95% variance Data: MIT-BIH Arrhythmia Database (10 records, 2,759 annotated heartbeats) Code: The complete implementation is available in the associated GitHub repository, including preprocessing,feature extraction, model training, and evaluation scripts. Keywords: Arrhythmia Detection, ECG Classification, Machine Learning, XGBoost, Signal Processing, HealthcareAI, MIT-BIH Database
Aqsa Shoukat (Wed,) conducted a other in Arrhythmia (n=2,759). End-to-end machine learning system (XGBoost) vs. Logistic Regression, SVM, Random Forest, and Neural Network was evaluated on Accuracy and F1-score. An XGBoost-based machine learning pipeline achieved 98.01% accuracy and an F1-score of 0.977 for arrhythmia detection using 2,759 heartbeats from the MIT-BIH database.