Feature selection for interpatient supervised heart beat classification

Area of Science:

• Biomedical Engineering
• Cardiology
• Machine Learning

Background:

• Accurate heart beat classification is vital for long-term cardiac monitoring.
• Current methods often use over 200 features, leading to suboptimal and complex models.
• Feature selection for ECG classification models is not consistently applied.

Purpose of the Study:

• To apply feature selection techniques to optimize feature subsets for ECG classification models.
• To evaluate the impact of feature selection on the performance of state-of-the-art ECG classifiers.
• To compare the performance of models with selected features against those with exhaustive feature sets.

Main Methods:

• Utilized feature selection techniques to identify optimal subsets of features.
• Evaluated classification model performance on real ambulatory ECG recordings.
• Compared results against previously reported feature choices for the same models.

Main Results:

• A small subset of individual features significantly contributes to accurate classification.
• Removing non-informative features leads to improved classification performance.
• Feature selection enhances the interpretability of ECG classification models.

Conclusions:

• Feature selection is a critical step for developing efficient and accurate ECG classification models.
• Optimizing feature subsets can overcome limitations of models using large, unselected feature sets.
• This approach offers a more interpretable and performant solution for cardiac function monitoring.