An Efficient Approach for Early Prediction of Sudden Cardiac Death Using Two-Stage Feature Selection and Gradient Boosting Classification

Sudden cardiac death (SCD) is one of the leading causes of death worldwide, resulting in unpredicted loss of heart function. This complex problem occurs in people with or without a history of cardiac illness. The symptoms of SCD start 1 h prior to its onset. The early detection of SCD may save many lives around the world. Hence it is vital to develop an accurate and precise method for identifying individuals at risk of developing SCD. This paper presents an efficient methodology for the early prediction of SCD using heart rate variability (HRV) and wavelet transform analysis by comparing diseased and non-diseased subjects. To accomplish this, the ECG signals of Normal sinus rhythm (NSR), Sudden cardiac death (SCD), and coronary artery disease (CAD) subjects were collected and pre-processed. HRV signals were derived from the ECG signal to extract various time domain, frequency domain, and non-linear method-based features. These features along with wavelet features and statistical features were considered for the selection of significant features. In this work, a two-stage feature selection method is proposed based on mutual information (MI) and recursive feature elimination (RFE) along with gradient boosting (GB) classification for accurately detecting SCD. Using the proposed MI-RFE-GB scheme, we achieved SCD detection 1 h before its onset with accuracy, sensitivity, specificity, and precision at 97.60%, 97.54%, 98.80%, and 97.59% respectively. The experimental results of the proposed scheme demonstrate the superiority over state-of-the-art methods. However, the current study can be extended using various cardiac disease datasets that cause for the development of SCD.