Ensemble Learning-Based Pulse Signal Recognition: Classification Model Development Study.

Area of Science:

• Cardiovascular signal processing
• Machine learning applications in healthcare
• Biomedical data analysis

Background:

• Traditional machine learning uses structured pulse signal data (time/time-frequency domains).
• Unstructured pulse signal data contains rich cardiovascular state information.
• Deep learning can extract local features from unstructured data.

Purpose of the Study:

• To fully exploit pulse signal information using both machine learning and deep learning.
• To compare classification performance of single classifiers versus ensemble methods.

Main Methods:

• Extracted structured data using time and time-frequency domain analyses.
• Employed Support Vector Machine (SVM) for structured data classification.
• Utilized Deep Convolutional Neural Network (DCNN) for unstructured data feature extraction and classification.
• Implemented a stacking ensemble method to fuse classification results.

Main Results:

• Single classifiers achieved a maximum average accuracy of 0.7914.
• The ensemble learning approach reached an average accuracy of 0.8330.
• Decision-level fusion of structured and unstructured data improved classification performance.

Conclusions:

• Ensemble learning effectively integrates information from both structured and unstructured pulse signal data.
• Decision-level fusion via ensemble methods significantly enhances classification accuracy.
• This study offers a novel, practical approach for objective pulse signal classification and diagnosis.