Unsupervised feature relevance analysis applied to improve ECG heartbeat clustering

Area of Science:

• Biomedical data analysis
• Machine learning in healthcare
• Physiological signal processing

Background:

• Computer-assisted analysis of biomedical records is crucial in clinical settings.
• Increasing data volume from modern devices challenges traditional data processing capabilities.
• Efficient data extraction methods are needed to manage large biomedical datasets.

Purpose of the Study:

• To develop an efficient data mining method for physiological records.
• To implement an unsupervised feature selection scheme for relevance analysis.
• To reduce the computational burden in analyzing large biomedical datasets.

Main Methods:

• An unsupervised feature selection scheme based on relevance analysis was developed.
• Least-squares optimization of the input feature matrix was employed in a single iteration.
• The algorithm generates a feature weighting vector to identify important features.

Main Results:

• The method was validated on electrocardiogram (ECG) records using a heartbeat clustering test.
• Achieved a 98.69% specificity, 85.88% sensitivity, and 95.04% general clustering performance.
• Reduced the number of features from an average of 100 to 18, with a 43% decrease in temporal cost.

Conclusions:

• The proposed feature selection method offers a highly efficient approach for biomedical data mining.
• This technique significantly enhances the performance of ECG clustering compared to existing studies.
• The method effectively addresses the challenge of processing large volumes of physiological data.