Feature Selection of Multichannel EEG for Attention Classification

EEG (electroencephalography) is a tool to determine human brain waves and the function of the human brain. A complete EEG device consists of 33 channels and is expensive for independent research. This study aimed to determine which EEG features play an important role in attention and meditation. Data were obtained from 83 respondents with 8 channels and 5 waves in each channel. This study examines three feature selection methods. Using Random Forest, selecting the Fp1_beta, Fp1_gamma, Fp2_beta, Fp2_gamma, Fz_beta, Fz_gamma, C4_beta, C4_gamma, O1_beta, O1_gamma, O2_alpha, O2_beta, O2_gamma features led to an accuracy of 98.2%. Selecting the Fp1_beta, Fp1_gamma, Fp2_beta, Fp2_gamma, Fz_delta, Fz_alpha, Fz_beta, Fz_gamma, C4_beta, O1_beta, O1_gamma, O2_alpha, O2_beta, O2_gamma features with a tree-based model led to an accuracy of 98.0%. Finally, with recursive feature elimination, the Fp1_beta, Fp2_beta, Fz_beta, O2_beta, and O2_gamma features led to an accuracy of 96.7%.