Abstract
In this study we propose a new machine learning classification method to distinguish brain activity patterns for healthy subjects. We used ElectroEncephaloGraphic (EEG) data associated with five userdefined mental tasks. We collected a data set using the Muse headband with four EEG electrodes (TP9, AF7, AF8, and TP10). Sixteen healthy subjects participated in this six-session experiment. In each session, we instructed them to conduct five different one-minute tasks, we abbreviated the tasks as think, count, recall, breathe and draw. After dealing with noise and outliers, we first fairly compared the performance of existing classifiers, including linear classifiers, non-linear Bayesian classifiers, nearest-neighbor classifiers, ensemble methods, and deep learning, with the same settings. Among these, Random Forest (RF), and Long Short-Term Memory (LSTM) outperform others. We then introduced a new ensemble classifier called Time Continuity Voting (TCV), combining these top two. The timewise cross-validation results showed that TCV could correctly classify the five tasks (20% by chance) with an accuracy of 70% which is at least 6% higher than the top individual classifiers.