Automatic sleep stage scoring system using genetic algorithms and neural network

We developed the automatic sleep stage scoring system using a 1-channel EEG signal. This system is a hybrid system constructed by the artificial neural network and the genetic algorithm using frequency and chaotic characteristics of an EEG signal. We designed the chromosomes with variable length and structure and fitness function to find the optimal input features and recognition network parameters. We applied the proposed method to 75 sample EEG signals for 5 sleep stages. The experimental results showed that the optimal sleep scoring system consisted of 3 layers with 15 to 30 nodes and found the optimal input features and initial weights to converge into the global minimum.