Table 1.
Demographics of patients in the sample (n = 788).
Table 2.
Statistics of features for 5-minute prediction.
Table 3.
Statistics of features for 10-second prediction.
Fig 1.
Plots of singular values and the principal components (PCs).
(A) Plot of singular values with the optimal hard threshold found by the method proposed by Gavish and Donoho.27 The singular values that are above the threshold are plotted in red while the ones that are below the threshold are plotted in black. The blue dot above the red dash line corresponds to the 14th singular value. The blue dot below the red dash line corresponds to the 50th singular value. (B) PC 14. (C) PC 50.
Fig 2.
First 18 principal components of HRV data matrix from one of the training sets.
Table 4.
Classification performance on test set for 5-minute and 10-second predictions using RF and SVM.
5-minute-pre-shock data contain 1000 R-R intervals while 10-second-pre-shock data contain 1600 R-R intervals. Each test was averaged over one hundred random cross-validation trials.
Fig 3.
Average importance of predictors for 5-minute prediction.
From predictor 1 to predictor 28: PC 1 –PC 18, α1, α2, mean N-N interval, power in 5 bins in frequency domain from low frequency to high frequency, Hjorth complexity and Hjorth mobility.
Fig 4.
Average importance of predictors for 10-second prediction.
From predictor 1 to predictor 28: PC 1 –PC 18, α1, α2, mean N-N interval, power in 5 bins in frequency domain from low frequency to high frequency, Hjorth complexity and Hjorth mobility.