Improved Support Vector Machine for ECG Signal Classification in Implantable Biomedical Devices

Electrocardiogram (ECG) signal classification in implantable biomedical devices plays a significant role for continuous heart monitoring and detecting cardiac abnormalities such as arrhythmias. These devices analyse ECG patterns to enable accurate diagnostics and provide necessary interventions. However, traditional ECG monitoring tools faces certain challenges like low power transfer efficiency, failure in capturing most paroxysmal Atrial Fibrillation (AFib), and difficulty in automatic determination of AFib. To solve this issue, the implantable ECG sensor with improved Support Vector Machine (SVM) is proposed to monitor the patient continuously and detect AFib effectively. The QRS complex is observed from ECG signals of 29 AFib patients is extracted by utilizing signal processing techniques such as wavelet transform and statistical analysis. The Minimum Description Length (MDL) method is applied to extract and select 12 features of variability-based heart rate parameters. Then, the improved SVM classifier with a Radial Basis Function (RBF) is used to classify the AFib in ECG data accurately. Experimental results demonstrate that the proposed implantable ECG sensor with optimized SVM achieves a better accuracy of 99.2%, sensitivity of 98.8%, specificity of 99.5%, and an F1-score of 98.9% which outperforms the existing methods like traditional SVM and CNN-based models.