Extreme Gradient Boosting Algorithm for Automatic Cardiac Arrhythmia Classification

Arrhythmia is a condition where the heart rate variability of an individual deviates from normal. Cardiologists cannot easily monitor the deviations. In a pandemic-like situation where the automated Classification of the disease can play a game-changing role, this paper presents extreme gradient boosting algorithm for automatic cardiac arrhythmia classification. In this study, heartbeats from the widely-used MIT-BIH database, consisting of 48 half-hour two-channel ECG recordings, are classified into one of the seven arrhythmia classes recommended by ANSI/AAMI. Time-frequency statistical features are extracted from the heartbeats using Wigner-Ville distribution coefficients, while frequency domain features are extracted from the third-order cumulant. All the coefficients of third-order cumulants are appended with statistical time-frequency features. A linear combination of the input features is extracted, which maximizes the separation between the classes by using Linear Discriminant Analysis (LDA), which is a supervised dimensionality reduction method. Machine learning algorithms used in this study are k-nearest neighbor (KNN), Support Vector Machine (SVM) and XGBoost. By using the XGBoost algorithm we obtained a score of 99.92% 99.21% 99.37% 99.29% and 99.88% for specificity, sensitivity, precision, f1-score, and accuracy, respectively. These results specify that the performance of the XGBoost algorithm is comparably similar to the other two algorithms. The results of this study suggest that the proposed model can be used in telemedicine environments even with much limited resources. The proposed system can be a significant breakthrough in pandemic situation providing a cardiologist level performance in classifying cardiac arrhythmia.