| ABSTRACT |
This present work aims to predict the sudden cardiac arrest (SCA) 5 min before the onset using HRV signal. This accuracy of predicting the SCA before its onset, sensitivity and specificity of the proposed algorithm is higher compared to the current state of art SCA prediction methods. This algorithm is based on HRV features derived from electrocardiogram (ECG) of sudden cardiac death (SCD) patients. We have used the two international standard databases (MIT/BIH - Arrhythmia Database for SCA prediction and Normal Sinus Rhythm (NSR) database for normal control) for developing this SCA prediction system. On SCD patients, the HRV signals of 1 min duration before 5 min of SCA onset is used to derive a set of statistical features on three different approaches namely, time domain, frequency domain and non-linear domain. There are 34 features were extracted from HRV signals of both normal and SCD patients. Feature reduction was done based on Sequential Forward Selection (SFS) algorithm and given to the machine learning algorithms (Fuzzy Subtractive Clustering (FSC), Neuro-Fuzzy Classifier (NFC) with Scaled Conjugate Gradient and Support Vector Machine (SVM)) for prediction task. The experimental results of this present study produced a maximum prediction accuracy of 94.74% using SVM and NFC classifier with 10 fold cross validation method. This prediction rate is higher than the earlier research works in the literature on the same database and also the SCA prediction time of the earlier works of 2 min duration is improved to 5 min on this work.
|
