Genetic Algorithm for Selection of Best Feature and Window Length for a Discriminate Pre-seizure and Normal State Classification

In the EEG based seizure prediction system, feature extraction and feature selection procedures which distinguish various states of the EEG signal are the main parts of the mentioned system. In the meantime, selection of appropriate window length for well discrimination of pre-seizure and normal states of the EEG signal is extremely significant. In this paper, a genetic algorithm based method was proposed for improving some dominant feature extraction parameters such as feature vector and its related window length. In this study, an appropriate representation of problem and fitness function for enhancing the described problem is selected. Eventually, we indicate that by applying these improved parameters, more discriminated classes -pre-seizure and normal classes -are obtained.