Feature Extraction from Electrocardiogram Data Based on Mathematical Modeling

• Published in: Quantitative bio-science pp. 119 - 129
• Main Authors: 서민경, 정대원
• Format: Journal Article
• Language: English
• Published: 자연과학연구소 01.11.2024
• Subjects: 통계학
• ISSN: 2288-1344; 2508-7185

This study proposes a novel approach for electrocardiogram (ECG) classification by leveraging the sparse identification of nonlinear dynamical systems (SINDy) algorithm for feature extraction. ECG signals represented as time series data are transformed into coefficients derived from second-order differential equations, enabling a compact and interpretable representation of the dynamics of each heartbeat. Using SINDy-extracted features, three machine learning models - logistic regression, random forest, and XGBoost - were evaluated on the MIT-BIH and PTB-XL datasets. For the MIT-BIH dataset, which involves multiclass classification, experiments were conducted on both the original imbalanced and resampled balanced data. Results indicate that SINDy-based features enabled high classification accuracy, with random forest achieving 97.39% accuracy on the balanced MIT-BIH data. In the binary classification experiments with the PTB-XL dataset, XGBoost achieved the highest accuracy of 92.98%. These findings suggest that differential-equation-based feature extraction offers a promising alternative to traditional ECG classification methods, combining high accuracy with interpretability. Future research will explore refining differential equations to capture complex ECG dynamics more precisely, as well as extending this approach to multicycle ECG patterns to further enhance classification accuracy. KCI Citation Count: 0