A systematic approach to seizure prediction using genetic and classifier based feature selection

Abstract

Currently, there is no standard approach for evaluating the intracranial encephalographic signals for seizure prediction. This study evaluates the IEEG signals by applying a systematic approach to feature selection, classification and validation to predict seizures. After preprocessing and processing, a genetic algorithm selects reasonable features off-line from a preselected group of features to serve as inputs to the classifier based feature selection process. A probabilistic neural network is used to select the optimal feature vector using a reed forward sequential approach on the training data followed by classification. A study of four patients resulted in a 62.5% average probability of prediction and a block false positive rate of 0.2775 false positive predictions per hour.