Fast feature selection algorithm of EEG data based on convolutional neural network

Electroencephalography (EEG) data is invaluable for investigating EEG physiology, neuroscience and clinical medicine. However, EEG signals are characterized by temporal nature, high dimensionality, and noise, which make it difficult for traditional feature selection methods for EEG data to extract useful information effectively with high computational effort. To solve this problem, this paper proposed a fast feature selection algorithm for multichannel EEG data based on a convolutional neural network (FFSEEG-GNN). Specifically, the algorithm pre-processes the multichannel EEG data, then extracts the waveform features and builds a CUDA architecture using GPUs to convolutional neural network operations. Finally, the most representative subset of features is selected in order of feature importance and feature extraction is performed. In addition, the simulation comparison experiment proves that the algorithm achieves 92.60% in terms of precision rate. Therefore, the program has accurate feature selection capability, which is expected to play an essential role in clinical medicine.