Prediction of resting-state EEG using high-dimensional patterns

Prediction of resting-state electroencephalography (EEG)usinghigh-dimensional pattern is a challenge due to the uniqueness of each person's brainwave. This study uses the headache EEG recording as the example, and predicts the informative different states by using an intelligent feature selection method. Vomiting and nausea are usually appeared in headache attacks, andit is sensitive to light, sound, or movement. In this study, we use the EEG recording with four classes (inter-headache, pre-headache, headache and post-headache) as the medical database. This study focuses three merits: First, we establish two balanced datasets which contain 2-class (inter-headache and headache) and 4-class brainwave datasets from the original imbalanced headache database so that there is no bias of the prediction system. The 2-class dataset consists of 22 subjects and 176 trials, and the 4-class dataset consists of 40 subjects and 320 trials. Secondly, we propose an efficient SVM-based method for predicting the headache attacks from the EEGby using an inheritable bi-objective combinatorial genetic algorithm (IBCGA). IBCGA automatically selects important features from the brainwave, and the 2-class prediction accuracy of leave-one-trial-out independenttest is 81.25%. Third, from the analysis of the brain region and channel frequency, the brain region T4 is the most important brain regions and alpha and beta frequencies are the most informative frequencies.