Forecasting Seizures in Dogs with Naturally Occurring Epilepsy
Seizures in epileptic patients can be anxiety-inducing, and the resulting medical and social issues can cause distress to a patient. A predictive mechanism to anticipate the onset of a seizure could help a patient prepare for a seizure and take medication. The literature has shown that the onset of a seizure is directly correlated with a distinct neurological change that can be detected using iEEG measurements. Thus the development of a classification model based on IEEG data can be created to aid epileptic patients. In this project we develop a feature set based on spectral and statistical features of the IEEG data, with and without ICA pre-processing of the data. Several classication algorithms were trained on the resulting data sets, including Naive-Bayes, SVM, logistic regression, and lasso-regularized logistic regression. It was found that the optimal results were given for regularized logistic regression on PIB features without ICA pre-processing, though generally all algorithms performed better than a random naive predictor model.