Machine Learning Approach to Recognition Human Mental State from EEG Signal

In this chapter, EEG signal analysis technique is examined for measuring mental stress of human being at different stages of attention. The goal of the current work is to identify discriminative EEG-based characteristics and suitable classification techniques that may classify brainwave patterns according to their intensity or frequency for the recognition of mental states that is beneficial for human–machine interaction. In this research work, three different mental states are classified such as relaxation, neutrality and concentration using Emotiv headset with four EEG Sensors (TP9, AF7, AF8 and TP10). In order to train and test various techniques, a dataset is developed for comprising five individuals and one-minute sessions for each category of mental attention. We tested a variety of features selection algorithms from an initial pool of 2100 features and application of multiple classifiers such as Bayesian Networks, Support Vector Machines and Random Forests allowed us to narrow the feature set down to only 44 critical factors to achieve an overall accuracy 87%.