Heartbeat Classification Based on Multifeature Combination and Stacking-DWKNN Algorithm

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning in Healthcare

Background:

• Arrhythmia poses a significant threat to human life, necessitating accurate diagnostic methods.
• Current arrhythmia classification strategies may lack sufficient accuracy for clinical decision-making.

Purpose of the Study:

• To propose an advanced classification strategy for distinguishing between normal and abnormal heartbeats.
• To enhance the accuracy and reliability of arrhythmia detection using a multifeature combination and a novel algorithm.

Main Methods:

• A four-module approach involving signal denoising, segmentation, and feature extraction from various heartbeat characteristics.
• Extraction of features including morphology, P, QRS, T length, intervals (PR, ST, QT, RR), and amplitudes (R, T).
• Utilization of the Stacking-DWKNN algorithm for classifying four types of heartbeats after optimal feature combination and normalization.

Main Results:

• Achieved an average accuracy of 99.01% on the MIT-BIH arrhythmia database.
• Demonstrated high sensitivity (89.42%) and positive predictive value (94.90%) for S-type beats.
• Showcased excellent sensitivity (97.21%) and positive predictive value (97.07%) for V-type beats.

Conclusions:

• The proposed multifeature combination and Stacking-DWKNN algorithm significantly improve arrhythmia classification accuracy.
• The method offers superior positive predictive value and sensitivity compared to existing models, crucial for clinical applications.
• This enhanced diagnostic capability supports more informed clinical decision-making in managing cardiac arrhythmias.