A genetic algorithm based time-frequency approach to a movement prediction task

The paper presents an investigation into a genetic algorithm based time-frequency approach for extracting features from the electroencephalogram (EEG) recorded from subjects performing a four-class self-paced movement task, left and right shoulder movement together with left and right foot movement. The objective is to predict left and right movements and accelerate the communication in a brain-computer interface (BCI). The features are attained by localizing the fast Fourier transformations (FFT) of the signals to specific windows localized in time. An interpolation approach is applied to reduce intra-class variations by smoothing the spectra for each signal. Some important parameters such as the overlaps of the FFT window, the points of interpolation during feature extraction are optimized by genetic algorithm (GA). The time-frequency features are classified by linear discriminant analysis (LDA). The approach achieves a good performance when quantified by classification accuracy (CA) rate, and has the potential to improve the performance of a brain-control based meal assistance system.