A machine learning approach for epileptic seizure prediction and early intervention

Epilepsy is often associated with modifications in autonomic nervous system, which usually precede the onset of seizures of several minutes. The identification of those changes is pivotal to predict the onset of seizure and to set up an early intervention. The aim of this study was to develop a patientspecific approach to predict seizures using electrocardiogram. Time- and frequency- domain features as well as recurrence quantification analysis variables, were extracted from the RR series. A machine learning approach based on support vector machine was then applied to predict seizures. The algorithm was applied in a dataset of 12 patients with 38 different types of seizures. A leave-one-out approach was used to predict seizures. An average sensibility of 83.8% and an average specificity of 72.8% were obtained. The results of the proposed approach show that it is feasible to predict seizure in advance, considering patient-specific, and possibly seizure-specific, characteristics.