Optimal EEG feature selection from average distance between events and non-events

Biosignal classification systems often have to deal with extraneous features, highly imbalanced datasets, and a low SNR. A robust feature selection/reduction method is a crucial step in this process. Sets of artificial data were generated to test a prototype EEG-based microsleep detection system, consisting of a combination of EEG and 2-s bursts of 15-Hz sinusoids of varied signal-to-noise ratios (SNRs) ranging from 16 to 0.03. The balance between events and non-events was varied between evenly balanced and highly imbalanced (e.g., events occurring only 2% of the time). Features were spectral estimates of various EEG bands (e.g., alpha band power) or ratios between them. A total of 34 features for each of the 16 channels yielded a total of 544 features. Five minutes of EEG from a total of eight subjects were used in the generation of the artificial data. Several feature reduction and classifier structures were investigated. Taking only a single feature corresponding to the maximum of average distance between events and non-events (ADEN) on unbalanced data yielded a phi correlation of 0.94 on the mock data with an SNR of 0.3, compared with a phi coefficient of 0.00 for principal component analysis (PCA). ADEN consistently outperformed alternative system configurations, independent of the classifier utilized. While ADEN´s high performance may be due to the nature of the artificial dataset, this simulation has demonstrated strong potential compared to other feature selection/reduction methods.