Fuzzy logic-based automatic alertness state classification using multi-channel EEG data

This paper represents an attempt to automatically classify alertness state using information extracted from multi-channel EEG. To reduce the amount of data and improve the performance, a channel selection method based on support vector machine (SVM) classifier has been performed. The features used for the EEG channel selection process and subsequently for alertness classification represent the energy values of the five EEG rhythms; namely δ, θ, α, β and γ. In order to identify the feature/channel combination that leads to the best alertness state classification performance, we used a fuzzy rule-based classification system (FRBCS) that utilizes differential evolution in constructing the rules. The results obtained using the FRBCS were found to be comparable to those of SVM but with the added advantage of revealing the rhythm/channel combination associated with each alertness state. © Springer-Verlag 2013