Publication: A Novel Hybrid Models For Identify Classification Of Ecg Abnormalities Using Dynamic Modal Decomposition And Machine Learning Approaches
Loading...
Date
2025-05
Authors
Liang, Chuchu
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Ecg signal classification is vital for diagnosing cardiovascular diseases. However, existing methods struggle with accurate classification due to challenges in extracting unstable modes, handling matrix rank mismatches, and processing time-series data. Traditional modal decomposition methods fail to capture unstable modes, losing critical features. This study proposes three models for ecg abnormality classification. The dmdc model combines dynamic modal decomposition (dmd) with 11 classifiers to extract unstable modes. The hdmd+svm model uses hankel matrix construction and singular value decomposition (90%) to detect abnormalities. The mdb-net model integrates dmd with bidirectional lstm networks (blstm) for optimized time-series processing. Experimental results show that dmdc improves accuracy by over 10% on the ptb database. Hdmd+svm enhances abnormal mode detection and robustness. Mdb-net achieves 87.3% accuracy and an auc of 0.9, outperforming other methods. These models address key challenges in ecg classification: dmdc focuses on unstable modes, hdmd resolves matrix rank mismatches, and mdb-net captures temporal features. Future work will optimize parameters and explore real-time ecg monitoring applications for early abnormality detection.
Description
Keywords
Machine Learning Machine Learning--Congresses Machine Learning--Educational Applications Machine Learning--Library Applications