Epileptic seizure prediction based on singular value decomposition of heart rate variability features

The present study proposes new method for epileptic seizure prediction based on heart rate variability (HRV) analysis and one-class support vector machines (SVM) technique. Methods: Excessive neural activity in preictal period affects not only brain activity, but also autonomic nervous system, that affects HRV. The proposed method distinguishes eight features, analyzes matrix of eigenvalues and eigenvectors, makes prediction based on one class SVM. Results: The method was tested on data collected from 31 patients. Total amount of preictal intervals is 232, interictal hours is 85. The result achieved are: sensitivity = 100%, specificity = 92%, precision = 92%, F1 = 92%, accuracy = 95%, fraction error length = 5%.