Using genetic algorithms and k-nearest neighbour for automatic frequency band selection for signal classification

The classification of signals is usually based on the extraction of various features that subsequently will be used as an input to a classifier. These features are extracted as a result of the experts´ prior knowledge, which may often involve a lack of the information necessary for an accurate classification in all cases. This study proposes a new technique, in which a genetic algorithm is used to automatically extract frequency-domain features from a set of signals, with no need of prior knowledge. This allows, first, to achieve greater accuracy in the classification of signals, and, secondly, to discover new data on the signals to be classified. This system was used to solve a well-known problem: classification of electroencephalogram (EEG) signals, and its results show a better performance in comparison with other works on the same problem.