Active learning methods for electrocardiographic signal classification.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Signal Processing

Background:

• Electrocardiographic (ECG) signal classification is vital for diagnosing cardiac conditions.
• Accurate classification often requires large, well-labeled datasets, which are time-consuming and expensive to create.
• Active learning offers a solution by strategically selecting informative samples for labeling.

Purpose of the Study:

• To introduce and evaluate three novel active learning strategies for ECG signal classification.
• To demonstrate the effectiveness of these strategies in improving classification accuracy with reduced labeled data.
• To compare the performance of margin sampling, posterior probability, and query by committee for ECG data.

Main Methods:

• Developed three active learning strategies: margin sampling, posterior probability, and query by committee.
• Integrated these strategies with Support Vector Machine (SVM) classifiers.
• Utilized both simulated ECG data and real-world ECG signals from the MIT-BIH arrhythmia database for experimentation.

Main Results:

• The proposed active learning strategies effectively identified significant ECG beat samples for labeling.
• These strategies demonstrated a promising ability to boost classification accuracy.
• A reduction in the number of required labeled samples was achieved while maintaining or improving performance.

Conclusions:

• Active learning is a viable and efficient approach for ECG signal classification.
• The presented strategies offer a practical method to build representative training sets with minimal manual labeling effort.
• These findings can lead to more efficient development of automated ECG diagnostic tools.